American Meat Science Association Research Guidelines for Cookery, Sensory Evaluation, and Instrumental Texture Measurements of Meat

Selection of samples

Researchers are strongly advised to consult with statisticians and/or perform appropriate statistical power tests for determination of experimental sample size and number of experimental units. A number of parameters, including an estimate of variability in texture (instrumental or sensory), flavor, or other traits of interest are needed to determine appropriate sample size. Many statistics textbooks provide guidelines for estimating an appropriate sample size to control Type II error. Likewise, there are many online options. It is more fiscally and scientifically sound to have adequate sample size than to use marginal numbers of observations and not detect differences that might truly exist (not controlling for Type II error). As a rule, regardless of the product or species being tested, all samples within a study should be representative of the products or processes under study. Products should be produced under the same conditions if possible. Proper sampling within muscles is of critical importance. Steaks or chops within a muscle that are to be assigned to different treatments should be randomized (if there is no known location variation) or blocked (if there is known location variation) to control for variance. Location can be accounted for either as random effects or fixed effects, depending on the design. If each carcass or cut represents 1 replicate of 1 treatment, steak location within a muscle should be standardized. For example, the first steak from the rib end of the short loin could be assigned to sensory analysis, and the second steak should be assigned to shear force evaluation tests. Some noncomminuted processed meats may need to be treated more like fresh meats when sourcing samples. For example, bacon would be randomized by belly and slice origin within the belly (a slice from the shoulder portion of the belly has different characteristics than a slice from the center or flank). A comminuted processed meat product, such as a sausage patty, would be randomized by and/or within production run/s.

Most research that has the goal of determining production, antemortem, or postmortem treatment effects on sensory traits utilizes a small number of muscles. Greater emphasis, however, has recently been placed on characterizing and marketing a greater number of muscles; therefore, it may be appropriate to evaluate treatment effects on multiple muscles. Location effects within muscle can contribute significantly to source of variation in some muscles. In addition, muscle selection should be based on experimental questions for the study. Data indicate that the relationships in tenderness among muscles in the same animal vary from relatively low to moderately high; therefore, treatment effects in one muscle may or may not be representative of similar treatment effects in other muscles (Ramsbottom et al., 1945; McKeith et al., 1985; Belew et al., 2003).

Processing conditions need to be taken into consideration when selecting processed meat samples. Multiple controls may be included if they are produced in the same plant and on the same line and equipment as other products in the study. Production should be in a steady state before any product is sampled for testing, and samples from a production shift should be pulled at 3 different times to ensure that the products are representative of the shift. The number of production dates or lots obtained will depend on the objective of the research. Standardization of the number of production lots and sample age are needed to accurately assess differences from product to product and should be representative of the product for which the results will be inferred.

Time postmortem for processing into cuts

Cuts should not be removed before rigor mortis is complete unless development of rigor mortis is an objective. The timing of rigor mortis completion may vary among species and with chilling conditions. Beef and lamb carcasses should not be ribbed and fabricated into cuts before 24 h postmortem, and pork carcasses should not be fabricated for approximately 18 h. Poultry and fish should follow recommendations based on completion of rigor. Timing is dependent on the research objectives and schedules of commercial facilities if the product is obtained commercially. For additional postmortem aging time, cuts should be vacuum packaged (unless dry aging is a component of research objectives). If a study includes alternative packaging systems in the experimental design, meat should be packaged to represent the objectives of the experiment. Extensive research has been conducted on the effects of aging on sensory properties of meat and meat products. Researchers are strongly advised to understand the effects of aging within their product prior to determining aging time/s for a study. The following postmortem times are recommended across species unless aging time is a factor of the study:

• •

  Beef: between 10 and 28 d, with 14 d being optimum for most research.

• •

  Pork: between 5 and 14 d, with 10 d being optimum for most research.

• •

  Lamb: between 7 and 14 d, with 7 d being optimum for most research.

• •

  Poultry: poultry meat deboning times vary. Two hours postmortem is common due to industry practices, 4 to 6 h postmortem or longer is recommended to detect differences between treatments that do not involve processing treatments, and aging poultry meat for 24 h postmortem prior to cooking or freezing is recommended.

• •

  Fish: due to variation in processing and slaughtering practices for commercially raised or wild-caught fish species, conditions including initial storage, freezing, and thawing, length of time for each respective stage, temperatures, and packaging should be defined and emulate commercial practices. Conditions should be standardized within a study.

• •

  Processed meats: 14 d postprocessing to allow for flavor equilibration.

These aging times are optimum for detecting differences among breeds and management systems, as well as antemortem and postmortem treatments without aging time functioning as a confounding factor, but they should not preclude research with unique postmortem treatments. Most aging effects on tenderness and flavor equilibration are complete by the above recommended times, but the industry average is considerably longer with wide variation within and among retail and foodservice products.

Meat dimension variables

Recommended thicknesses of steaks, chops, filets, patties, and processed meats are as follows: beef steaks (dry heat): 2.54 cm; beef steaks (moist heat): 1.9 to 2.54 cm; lamb chops (dry heat): 2.54 cm; and pork chops (dry heat): 2.54 cm. Whole poultry filets should be at least 2.54 cm in thickness or greater to represent the natural thickness of the filet. Due to portioning filets, samples may be thinner if product is marketed sliced such as thin-sliced chicken breasts, and a standard thickness should be defined.

Fish filet thickness depends on species, and for larger species, thickness should be defined based on common use of the product. Beef patties (also recommended for pork and poultry) should be between 0.95 and 1.10-cm thick for 91.5-g patties and 1.10 to 1.27 cm for 113.5-g patties. For specialty beef patties, thickness should be defined by product type such as smashed patties or hand-formed patties, and processed meats should be representative of how they would be utilized for foodservice or consumer preparation.

These are recommendations, and thinner or thicker products can be used as long as thickness is standardized within the study. If products are too thin, it is difficult to accurately measure the internal temperature during cooking. Cut thickness influences heat transfer, protein and lipid heat denaturation, and moisture loss, all of which have subsequent effects on sensory properties.

A cutting guide should be used so that meat products are uniform in thickness. Commercial meat slicers also can be used to generate fresh product from carcasses or cuts. When the product must be frozen, uniform thickness can be obtained by freezing and cutting with a saw or slicer.

Because it is impractical to list all the various roast cuts from all species, the latest editions of the Meat Buyer’s Guide (Meat Institute, 2025) and the Institutional Meat Purchase Specifications (USDA, 2014) provide common sizes of products merchandized in retail and foodservice outlets. Before cooking, retail cuts should be trimmed of excess external fat. The following are minimum weights and thicknesses for roasts: beef:1.5 kg, 5.0-cm thick; pork: 1.0 kg, 5.0-cm thick; lamb: 0.5 kg, 5.0-cm thick; and poultry: either whole, halved, or quartered carcasses as defined in the experimental design.

Freezing and frozen storage

For instrumental and/or sensory testing, steaks, chops, filets, or processed meats should be cooked, sheared, or evaluated for texture without freezing whenever possible. Freezing and thawing has been shown to decrease shear force compared to the fresh/never-frozen values for shear force due to muscle-fiber damage and ice crystal formation (Wheeler et al., 1990; Grayson et al., 2014; Beyer et al., 2024), but this can be minimalized with rapid, quick freezing and thawing at refrigerated temperature (0–4°C). When the experimental design is such that sample size or treatment protocol requires freezing of samples (multiple days of aging for example), it is acceptable as long as freezing occurs under standardized conditions across the experiment and the potential impact of freezing on results is acknowledged. As freezing samples at different aging times results in unequal frozen storage durations across treatments, these effects can be minimized by standardizing freezing and storage conditions and by randomizing aging days across sensory testing days so that each aging time is represented within a testing session.

Freezing should occur so that all samples freeze in the same amount of time by spreading samples out in one layer rather than placing them in a box and then placing them into the freezer. Likewise, rapid freezing at very cold temperatures, preferably lower than −20°C, to limit damage from ice crystal formation is recommended (Hiner et al., 1945; Qian et al., 2022). Pork and ground beef patties should be evaluated within 3 mo of frozen storage; beef and lamb steaks, roasts, and chops should be evaluated within 6 mo of frozen storage; and poultry and fish should be evaluated within 3 mo of frozen storage (AMSA, 2016). The purpose of recommended frozen storage times is to minimize lipid oxidation. Additionally, frozen sensory samples should be vacuum packaged with a high oxygen barrier film to assist in delaying lipid oxidation effects during storage.

The same freezing process recommendations apply to processed meats. The length of time that processed meats should be stored in frozen conditions depends on the characteristics of the individual product and the objective/s of the research. For example, if the researcher seeks to understand the impact of a new antioxidant, the test should be extended through the entire shelf-life of the product. In contrast, if a new flavor of sauced chicken wings is being developed, it is generally recommended to not test after 50% of the shelf-life of the product has occurred or when recognized changes in flavor and/or texture are detected. If the objective is to determine shelf-life, then an appropriate longer time would be used. For example, if the shelf-life of a fresh sausage product is approximately 17 d, you want to evaluate the product for a minimum of 17 d. Microbial level and potential pathogens should be assessed prior to sensory testing for food safety reasons.

Physical state and packaging of the meat product prior to cooking

Fresh meats, steaks, chops, or filets should be vacuum packaged in oxygen-impermeable film to avoid desiccation and oxidation. Internal temperature of the meat at the initiation of cooking can affect tenderness (Berry and Leddy, 1990; Wheeler et al., 1997b). Likewise, if the temperature is lower than 2°C, the cut is prone to surface charring. The internal temperature should be recorded just prior to placing meat in the cooking environment. It is acceptable to cook ground beef patties from fresh, frozen, or thawed, depending on the objectives of the experiment as long as conditions are standardized.

Thawing conditions should be standardized for temperature, packaging, and length of time. Thawing should never be at room temperature or in water where water temperatures are above 5°C. It is preferred that raw or fresh meat products be vacuum packaged during thawing to reduce the effects of lipid oxidation. Products should be thawed in a refrigerated room with a temperature of 2 to 4°C and spaced to not overlap. The length of time for thawing should be standardized such that the internal temperature of the product reaches 2 to 4°C. Preliminary tests should be conducted to determine length of time, usually 24 or up to 48 to 72 h for larger samples. For example, vacuum-packaged beef, pork, and lamb or turkey roasts should be thawed at 2 to 4°C for 48 to 72 h or the length of time required to reach an internal temperature of 2 to 4°C. Frozen, thawed, and cooked weights should be collected for the determination of losses at different stages from the frozen to the cooked state and may include purge loss after thawing or during storage.

For processed meats, products should be treated as typically defined by the manufacturer and the objectives of the experiment for packaging, storage, and preparation (thawing). The decision to cook from the refrigerated, frozen or thawed state should be based on the product and experimental objectives. Regardless of state, temperature should be standardized at the initiation of the cooking process.

Typical storage practices for various processed meat products are presented in Table 1. Typically, products should be cooked from the intended storage state. For example, retail fried chicken products, such as chicken nuggets, are typically stored frozen in a retail plastic bag and cooked from the frozen state. Frankfurters are typically stored in a vacuum-packaged bag in refrigerated conditions, then cooked from the refrigerated state. On the other hand, premarinated individually quick-frozen poultry products may be thawed prior to cooking, and shelf-stable dried sausages can be stored and served at room temperature. Frozen, thawed, and/or cooked weights should be collected for the determination of losses at different stages from the frozen to the cooked state (if applicable).

Methods for monitoring temperature changes

Throughout the cooking process, it is imperative to monitor product and equipment temperature to prevent under, over, or uneven cooking. Both the initial and endpoint product temperatures, cooking time, equipment temperature, and product-evaluation temperature should be recorded to maintain and document consistency across samples. Temperature monitoring can be conducted using a thermocouple to determine internal meat temperature. Infrared guns or metal grill temperature monitors can be used to monitor the surface temperatures of equipment or meat exterior surfaces. Prior to use, either device should be calibrated to 0°C using ice water and to water at 100°C or appropriate boiling temperature based on altitude. Thermocouple use recommendations are presented in Figure 2. When using infrared guns, distance from the object, angle held, and environment may impact readings. The gun should be held at the manufacturer’s recommended distance from the surface at a 90° angle. Steam emanating from the surfaces may influence readings, and metal oven temperature monitors may be more accurate.

Figure 2.

Procedures for measuring internal meat temperature.

Evaluating degree of doneness

Because of the concern regarding food safety and for consistency in cooking that provides the greatest likelihood of detecting treatment differences, all meat in an experiment should be cooked to the same temperature, unless the effects of endpoint temperature are part of the experiment. Researchers should collect information regarding cooking times, peak/endpoint temperature, and cooking yields. Because of differences in cooking rate, the method of measuring “cooked temperature” is important. Some researchers prefer cooking to an endpoint temperature and then removing the samples from the heat, whereas others prefer to remove samples from the heat before the target endpoint is reached and then monitor temperature rise and record the peak temperature (usually for cooking methods with a high cooking rate or for heavier weight cuts). It is recommended, if cooking at a lower rate such that the postcooking temperature rise is less than 2°C, meat should be removed from the heat at the endpoint temperature. If cooking at a high rate such that the postcooking temperature rise is greater than 5°C, then determine at what temperature or cooking time samples should be removed from the heat to obtain the target endpoint temperature after the postcooking temperature rise (Figure 2). Appropriate thermocouples and recorders should be used for measuring and recording internal temperature.

Changes in color, interpreted as degree of doneness upon completion of cooking, may be influenced by many factors, including animal maturity, type of muscle, fat content, added ingredients, method of cooking, pH, length of cooking, and internal temperature. Color variation in beef patties, such as “premature browning” or “persistent pinking” when cooked to constant final temperatures, poses a unique problem regarding safe final internal temperatures based on USDA (2025). Because of the possibility of color variation in beef patties, degree of doneness based on cooked color cannot be used as a reliable indicator of the selected temperature endpoint; therefore, the actual temperature should be measured. For poultry, the actual endpoint temperature must be measured.

Roasting

Roasting is a method by which heat is transmitted to the meat by convection, either by normal or forced air, in a closed, preheated oven. As indicated above, forced-air convection-oven cookery is not recommended. For traditional roasting, meat is placed on a rack either in or over a shallow pan to catch drippings. The oven door is closed, and the meat is not turned during cooking. For rotisserie-style roasting, meat is placed on a rotating spit. The oven door is closed, and the meat rotates during cooking. If cooking multiple treatments at one time, ensure that the drippings do not contaminate the other treatments. Roasting procedures are reported in Figure 3.

Figure 3.

Cooking procedures for roasting meat.

Broiling

Broiling is cooking meat using direct radiant heat, which is also referred to as a “broiler.” The meat is placed above or below the heat source. The heat usually radiates from one direction, so the meat must be turned during cooking. In these guidelines, this method will be referred to as “broiling.” Many investigators prefer broiling because it closely resembles methodology that is commonly used by consumers. Others prefer the “oven-roasting” technique because the cooking conditions can be more easily controlled. Cross et al. (1979) compared roasting and broiling techniques for beef steaks and indicated that both methods were acceptable and resulted in steaks with similar sensory characteristics. Figures 4 and 5 define broiling and belt-grill cookery procedures for meat, respectively.

Figure 4.

Cooking procedures for broiling meat.

Figure 5.

Cooking procedures for belt-grill cooking of meat.

Pan broiling

Pan broiling is a method by which small, thin cuts such as patties or bacon are cooked by direct heat of conduction. Meats are placed in frying pans or electric griddles and are cooked without added fat or water. Meats are turned frequently if needed to prevent excess surface browning or crust formation and to allow for more even cooking. The pan or griddle should not be covered during pan broiling. Fat or drippings should be drained between samples to prevent uneven degrees of doneness, cross-contamination across samples, and prevent lipid oxidation. The cooking procedures are presented in Figure 6.

Figure 6.

Cooking procedures for pan broiling meat.

Impingement oven

Impingement ovens use air heated by gas or electricity that is forced into a chamber or chambers (in large commercial systems, multiple chambers are used to control the cooking method), and the product is carried via a belt, usually a wire mesh belt, through the chamber. Some systems have humidity controls within each chamber. These systems have replaced belt grills for some institutions.

Humidity, room temperature, steak, chop, filet, or patty thickness, and factors influencing heat transfer within the product affect dwell time that is necessary during impingement cooking to reach a defined internal temperature. Different dwell times are expected across days and within a day, and therefore, dwell time and internal endpoint temperature should be monitored routinely within a day and reestablished each cooking day. Impingement ovens are also frequently used to prepare processed meats, specifically pizza toppings (i.e., pepperoni, sausage crumbles, etc.). When evaluating pizza toppings, consider evaluating independently as well as in application, depending on the research objective/s. Dwell time and chamber temperature should be predetermined based on preliminary testing. During preliminary testing, select the dwell time and chamber temperature that produces the appropriate visual and textural degree of doneness of the pizza topping and/or an entire pizza. Methods are presented in Figure 7 for impingement oven cooking.

Figure 7.

Procedures for cooking meat in an impingement oven.

Clam-shell grilling

Clam-shell cookery simulates grilling by using either a flat griddle or raised grill-mark heated surface with a lid that closes on top of the sample. The top and bottom surfaces are heated simultaneously. This is a popular method of cooking that has largely replaced open-hearth electric grills. Clam-shell cooking procedures are reported in Figure 8. Clam-shell grills may have temperature variations across both surfaces. Preliminary studies should be conducted to determine surface temperature variation. Also, the amount of time needed for a grill—both top and bottom—to return to the predetermined surface temperature after opening and removing a cut, needs to be defined. For grills that automatically turn off, protocols are needed so that this function does not affect cooking or induce variability in meat samples that is not due to treatment effects.

Figure 8.

Meat cooking procedures for clam-shell grills.

Baking

Baking uses dry heat in an oven from either an electric or gas heat source without air circulation. While oven temperatures can range from 90 to 250°C, meat is usually baked at 175°C. During baking, meat products are either covered or wrapped in foil to provide a moist environment. Baking is commonly used for poultry, fish, and further-processed meat products. Procedures for baking are presented in Figure 9.

Figure 9.

Meat cooking procedure for baking.

Deep frying

Deep frying involves submerging food in heated shortening or oil. Deep frying is a popular method for cooking breaded meat items. Deep frying, especially for food service products, should be carried out using commercial frying equipment; however, retail frying equipment can be used if necessary. Frying time, oil temperature, oil type, and product load should be predetermined based on preliminary testing. This will vary depending on the product and individual fryer. The oil should be filtered and changed according to oil quality measures to minimize off-flavor migration from the cooking oil. Depending on the treatment, researchers should consider variability between fryers that may contribute to significant sources of variation. During preliminary testing, determine settings that will result in product that meets the required internal temperature and desired breading characteristics. Procedures for deep-frying cooking are presented in Figure 10.

Figure 10.

Deep-frying cooking procedures for meat.

Pan frying

Pan frying uses a pan placed on a heat source or an electric skillet where a standard amount of fat or oil is added to aid in heat transfer, reduce the adhesion of meat to the pan, and to impart a minimal amount of flavor. Pan-fried meat products are cooked by direct heat of conduction and heat of the added fat. The pan is preheated to a standard temperature, and the preweighed or measured fat or oil is added. Meats are then placed in frying pans or electric skillets. Internal temperature is monitored, as previously defined, if the meat product is of sufficient thickness. If the product is too thin to insert a thermocouple, preliminary studies can be conducted to determine time of cooking. Meat products should be turned over one time during cooking at approximately the midpoint of the cooking process. The pan or skillet should not be covered during pan frying. Fat or drippings should be drained and cleaned from the pan between samples to prevent cross-contamination of flavors across samples and to assure that the fat source is fresh and the same for each product that is cooked. Pan-frying procedures are presented in Figure 11.

Figure 11.

Cooking procedure for pan frying and sautéing meat.

Sauté

Sautéing includes the addition of flour to the product prior to cooking and uses cooking methods similar to pan frying. As indicated in Figure 11, sautéed meat is dusted or dredged in flour prior to being placed in a pan containing a standard amount of oil. The addition of flour provides a coating during cooking and may affect flavor and cooking parameters of the meat. The weight of the sample before and after the addition of flour is needed to account for flour pickup. The remaining steps for sautéing are the same as those listed for pan frying.

Steeping

Some meat products, such as frankfurters, should be reheated by steeping. A standard amount of water is placed in a pan, the pan is placed on a heating unit, and the water is heated to a standard temperature, which is usually boiling. The product is then placed in the water with the water covering the product. The pan and product are removed from the heat source, covered on top, and allowed to sit in the heated water for a defined amount of time. The amount of water to be added to the pan and the product load should be defined in preliminary studies. The final cook temperature is determined at the end of the steeping time. The steeping time should be reflective of recommended times defined by product manufacturers. Figure 12 defines procedures for steeping meat products. It is important to clarify that steeping and/or boiling should not be used in the evaluation of fresh meat products.

Figure 12.

Procedures for steeping meat products.

Sous-vide

Sous-vide is a moist cooking method where the product is placed in a heat-stable vacuum-packaged bag and cooked in water (Hrdina-Dubsky, 1989; Church and Parsons, 1993; Creed and Reeve, 1998). Immersion water-circulating systems are used to maintain a predetermined standardized water temperature. Water is circulated by the system to maintain constant water temperature and even temperature exposure by the vacuum-packaged meat sample. The size of the water container and product load per cooking cycle need to be standardized. Water cook temperatures are lower than cook temperatures used in the aforementioned methods and usually are 80°C or less, depending on the product application. Cooking and endpoint temperatures vary in industrial applications across products. Researchers are encouraged to define and emulate industry temperatures and practices. Meat products can be fully or partially cooked using this system. Preliminary studies to determine water temperature, circulation rate, approximate length of cooking time, and internal endpoint temperature need to be determined. Internal product temperature can be monitored by placing a thermocouple in the geometric center of the piece of meat by piercing and threading it through the surface of the vacuum-packaged bag using a preformed, water-proof septum or tape. After cooking, the product is rapidly cooled and then reheated for serving to eat after a period of chilled storage at 1 to 4°C (Creed and Reeve, 1998).

This cooking process reduces cook loss and affects product flavor and texture (Vaudagna et al., 2002; Cabral, 2019; Latoch et al., 2023). With this cooking method, surface dehydration and development of the Maillard reaction during cooking is very limited or does not occur. With the reduced presence of Maillard reaction products and surface browning, sous-vide products are often subjected to a rapid surface heat treatment to obtain brown color on the surface and to allow for the development of Maillard reaction products. The rapid heat treatment can be roasting, broiling, pan broiling, pan frying, or any of the other aforementioned methods that use high heat. Parameters of the reheating process and final endpoint temperature should be recorded. It should be noted that when water temperatures below 62.8°C are used, there are concerns with inactivation of bacteria (Hunt et al., 2023). The texture of sous-vide meat cooked meat differs from that of other cooking methods (Cabral, 2019; Gil et al., 2023; Latoch et al., 2023). Figure 13 defines cooking procedures for sous-vide cooking.

Figure 13.

Cooking procedures for sous-vide cooking of meat.

Sensory testing environment

The methods used for preparing and presenting samples to the panel require decisions about the testing environment, number of sessions, and physical condition of the samples. The validity of sensory panel results is partially dependent on the control of various factors within the testing environment. Generally recommended procedures for panel room location and layout, lighting, odor control, and comfort of panelists are presented in ASTM MNL26-3^RD-EB (2020) and Meilgaard et al. (2025). Detailed recommendations on these items as well as sensory panel booth design can be found in ASTM MNL60 (2016).

The sensory testing facilities should be in an accessible location with sufficient space, temperature and humidity control, and freedom from noise and odors. There should be a slight positive pressure within the sensory room to prevent cooking odors from entering the room. Sufficient electrical outlets and Ethernet or Wi-Fi access should be available for direct data entry. The room/s should be free of other distractions, such as photos or posters on the walls, overly distracting paint colors or wallpaper, or obvious product brand name placement or identification.

Sensory panel booths are commonly used for product evaluations by both trained panels and consumers, as they maximize the ability to control the testing environment (lighting, temperature, food odors, noise, etc.). If space is at a premium, these booths can be portable and capable of quick assembly. Incandescent, fluorescent, or both incandescent and fluorescent lighting can be used in the testing area. Whichever approach is chosen, the lighting should be uniform. The use of a dimmer switch is desirable because it allows for a variety of intensities ranging from 70- to 80-ft candles, which is typical in an office area, to higher levels of 110-ft candles. The light intensity should be measured, reported, and measured across testing days to ensure that uniform lighting conditions are present for each sensory session within a study. If darker lighting is used, differences in appearance may be partially masked or not be as apparent. For any sensory test, visual or oral, light intensity, either foot candles or lux, should be measured and reported.

Color differences may occasionally occur in a study that are not related to the variables being tested and therefore need to be masked. Colored lights (bulbs or filters) can be used in taste panel booths to mask these color differences. Shaded glasses or goggles may also be used; however, Sipos et al. (2021) reported that panelists can easily look around the shaded glasses or move the goggles aside to view the samples. A benefit of adjusted room lighting is a more complete light saturation of the testing environment. Red lights may be most effective in masking variations in reds, pinks, and different degrees of doneness. Amber-colored lights may effectively mask variations in light brown and tan colors. Preliminary studies should be done to determine which light color masks the color differences most effectively. Consider dimming the lights in the sample preparation area so that the light is not transmitted to the booths when the pass-through door is opened.

When testing with consumers, however, colored lights should only be used if absolutely necessary because colored lights can cause artifacts or nontypical responses from consumer panelists. The colored lights might make them suspicious and more attentive to smaller differences in the samples that they would not normally detect using standard lighting. When testing meat products that may be cooked to varying degrees of doneness, consumer panelists whose preferences match the degree of doneness may be preferred. However, if testing the effects of degree of doneness across consumer preferences, understanding consumer degree of doneness preference may assist in interpretation of the results.

Panelists for descriptive analysis panels may be seated around a table, usually a round table, or evaluate individually in booths. Panelists who have extensive training and experience should be able to independently evaluate products using a round table without the biases associated with using untrained panelists. The panel leader should assess the panel’s level of training and only place panelists around a table when they are ensured that this environment will not result in communication and influence of one panelist to others. For consensus data collection, a round table is recommended to allow for discussion.

Additional options for testing locations

While a sensory laboratory is the best location for testing with trained panelists, because of the ability to control the testing environment, there are several alternative options for testing locations when conducting consumer tests. The choice of testing location depends on the test objectives and the compromises that researchers are willing to or must make. The ability to control sample preparation and test administration, which includes noise and distractions, varies greatly depending on the type of testing location that is chosen. Field agency facilities are similar to a sensory laboratory where there is a greater ability to control sample preparation and test administration. To improve testing efficiency, panelists are often seated at individual tables in a classroom-style arrangement with all panelists facing the same direction. School cafeterias and church facilities can also be used, as they provide accessible venues with existing populations for recruitment and allow demographic characteristics to be managed through selection of locations and participant groups within each city. There is less ability to control the atmosphere and distractions in these environments. Testing in school cafeterias, retail stores, mall intercepts, and restaurants have similar issues, including excessive amounts of conflicting aromas, noise, and distractions. The panelist sample size should, therefore, be increased to accommodate the less-controlled testing environment. The number of product samples served should also be minimized as shoppers’ attention is compromised. When using more uncontrolled locations, standard errors would be higher when calculating the recommended sample size when compared to controlled environments. In-home testing is the location that has the least amount of testing control, but it emulates actual use conditions. This is important in testing packaging features, storage, and ease of product preparation.

Consumer testing locations should be vetted and equipped to handle the type of testing required and the products evaluated. Key considerations include adequate kitchen space and the necessary equipment. If the proper equipment is not available, it should be installed. The preparation area should be far enough away from the evaluation area to prevent cross-contamination of odors and the noise distraction. Proper transfer of equipment should be available for bringing prepared samples to the testing area. Storage space (dry, frozen, or refrigerated) for all products, serving equipment, tasting materials, etc. should be available. If using frozen or refrigerated storage, product temperature should be monitored via electronic or manual checks.

Enough staff should be retained to sign-in participants, prepare and serve samples, monitor the testing, and provide payment once testing is complete. The ability to collect and handle data in a robust way is imperative. The facility should have procedures in place for collecting and storing data. If testing in the home, ballots or online surveys should be straightforward and easy to understand, and staff should be available by phone to answer additional questions. The facility should be located near consumers and be able to recruit consumer group/s of interest from the area/s that the product is typically purchased and consumed. For example, consumers of a product not available in the Northwest should not be recruited by a facility in Portland, Oregon.

Preparation and serving size of sensory samples

Selection of sample preparation method and serving size should be determined based on project objectives and the amount of variation between and within treatments. It is critical that each panelist receives a standardized amount of each sample. If there are multiple preparation methods that consumers typically use in preparing the product being tested, the following factors should be considered:

• •

  Which preparation method is used most often?

• •

  Which method yields the best sensitivity and consistent product quality?

• •

  Which method is more likely to uncover or result in detectable differences?

The cooking method that yields the most sensitivity and consistency may not be the method used most often. If the cooking method results in a less consistent product, the resulting higher degree of variability may cause key sensory differences to be missed due to the loss of test sensitivity. Careful testing should be done to ensure that the cooking method that is selected yields similar sensory properties to the more commonly used cooking method/s. Alternatively, consider including multiple preparation methods if needed.

Regardless of which cooking method is used, it is critical that each panelist receive a standardized amount of each sample. To reduce the chances of introducing variability, samples should be standardized based on weight or dimensions. Serving size should be based on the test objectives, whether the panelists are trained or untrained consumers, and the number of questions or attributes to be evaluated. It is recommended to serve enough of each sample to provide an accurate assessment of the sample and account for inherent variability but not too much to cause taste bud fatigue. Depending on the test objectives, the serving size could be based on the serving size listed on the package label, otherwise use a representative serving size that accounts for the inherent variability in the product. It is recommended to discuss sample preparation and presentation with a sensory professional to develop the best preparation and presentation design of the meat sample.

To account for the moderate to sometimes high degree of variability among and within treatments, meat samples often are cut into cubes, and each panelist receives 2 to 3 cubes from different locations within the piece of meat. For steaks, chops, and roasts, cubes that are 1.27 cm × 1.27 cm × thickness of the cooked cut (2.54 cm) are suggested. If cooking procedures result in variation in cut thickness and charred surfaces, however, the thickness dimension should be standardized and cooked surfaces removed from cuts. Removal of cooked surfaces should be carefully considered, as the influence of cut surface on flavor and texture may be a desirable aspect of the sensory attributes of a sample or may be a component of the sample’s sensory properties. To cut cubes, it is recommended to use a sample sizer. After the hot sample is trimmed of all bone and epimysial (outer) connective tissue, it is placed in the plexiglass sample sizer (Figure 14). The sample sizer should have dimensions of 14-cm long × 12-cm wide × 4-cm deep to accommodate large cuts. On each side, the slots are spaced 1.27 cm apart and have an opening of 3 mm to allow the knife to cut the sample in each direction. For beef patties (depending on the size of beef patties being evaluated [91.5 or 113.5 g]), cooked patties can be cut into 6 or 8 pie-shaped samples as shown in Figure 15. Even with thicker or larger-sized patties, cutting patties into cubes might result in breakage and the inability to obtain equal-sized pieces to serve the panelists. Cutting patties into pie-shaped or wedge samples is recommended.

Figure 14.

Sensory sample sizer for steaks or chops.

Figure 15.

Sectioning beef patties for sensory evaluation.

Protocols followed by a trained panel evaluation should be clearly documented to allow for a repeatable and consistent evaluation. During the normal eating experience, consumers assess the juiciness of the meat visually as well as get an initial impression of the tenderness as they cut the sample into bite-sized pieces. Therefore, when conducting consumer tests, it is best to serve samples that are large enough for the panelist to cut to provide a more accurate representation of the actual consumer eating experience. The serving size should be standardized. Location effects within a subprimal should be randomized. The effect of location can then be included in the analysis of variance (ANOVA) model, usually as a random effect.

Depending on the objectives and circumstances of the study, there will be times when it is advisable to serve the smaller 1.27-cm cubes. By randomizing the selection of cubes within a sample, the variability in tenderness within a sample can be better accommodated. These results, however, are a poorer simulation of the normal eating experience when panelists receive bite-sized cubes. Researchers should understand that they are giving up aspects of the eating experience that influence consumer perception when serving 1.27-cm cubes, and therefore, the interpretation and use of the results might be affected. The experimental design might need to be altered (i.e., increase the number of consumers per treatment) if cubes are used.

For poultry, meat and similar products that have been shown to vary significantly by location across the sample, serving size, methods, and evaluation procedures need to be discussed and determined. After preparation, it should be determined what sample size is large enough to account for variability within treatment. For trained panel evaluations, it may be necessary to evaluate poultry pieces independently, such as the head vs. tail portions of the breast portion, thigh vs. drumstick, and so forth. Panelists may be presented with either standardized pieces from each section or a whole chicken breast. If a whole chicken breast is presented, panelists would be instructed to cut the breast in half and eat from the center and remove both the anterior and posterior from the center cut. Panelists would rate individual areas and an overall impression of each attribute. For consumer evaluation, the head and tail ends of the breast should be cut off and discarded. The center portion of the breast should be cut in half and served either randomly, or consumers can receive the same breast portion of the center section of all samples in the study. If samples are randomized by portion, tail or head, these data should be identified and included in the statistical model. Note that with randomizing samples by location, a greater number of consumers may be needed in the study.

Samples for trained panel evaluations for processed meats are typically served by themselves (without condiments or bread). As the objective is to determine if differences are detectable, a greater amount of test sensitivity is desirable. However, when conducting consumer acceptance tests, whether to serve the sample by itself or “in application” (i.e., as a sandwich or on a bun and/or with condiments) should be considered. Serving the sample in application better replicates the actual consumer experience. The risks and benefits of serving in application should be carefully considered, as serving in application introduces variability and may result in fewer differences being detected.

Minimum recommended serving sizes for various processed meat products and factors that should be considered when designing research are presented in Table 2. The study objective, test method, length of questionnaire, and number of samples should guide the actual serving size. It is imperative that sample size be standardized across all servings.

When evaluating fish whole-muscle products, variation within the filet or in the whole sample has to be accounted for within the sensory sampling. As fish may vary in harvest size, the number of samples or experimental units selected to represent a batch, either wild-caught or farm-raised fish, should be considered. As there is most likely more variation in wild-caught fish, batches usually defined as a catch within a fishing day and selection of fish as experimental units within a batch that are representative of commercial size and weight should be defined. Filet thickness can vary, and there may be variation in texture across the filet. Accounting for variation within a whole-muscle sample, such as discussed for chicken breasts, should be included in the experimental design. Seafood products where the whole piece provides a portion, such as shrimp, clams, oysters, and scallops, should be sampled randomly from a batch. Sufficient product should be presented for trained or consumer evaluation to account for variation within the batch and to provide for at least 2 to 3 portions. In studies where panelists are asked to evaluate multiple attributes, serving portions should be sufficient for evaluating all attributes. For processed seafood products, considerations as defined for processed products across other species are similar.

Sample presentation

Standardized presentation procedures need to be followed to minimize the introduction of extraneous variability. This includes standardizing the critical test elements such as serving temperature and holding conditions, the number of samples per session, and the order of sample presentation.

Serving temperature and holding conditions

Samples should be served at the most appropriate temperature for the attributes being measured. Serving temperature must be achievable and be able to be maintained consistently yet allow for any sensory differences to be detected. Temperature should be monitored to ensure that the samples are a standard temperature when serving and that the temperature is not too high or low. Variation in temperature of samples when the panelists receive them should be documented. As flavor and texture are affected by temperature and holding time, not only should the serving temperature be consistent, but the holding time of the cooked samples should be consistent as well. For each study, procedures for holding, cutting, and serving should be determined ahead of the study.

ASTM Standard E1871-17 (2017) recommends that meat and poultry samples should be served at a minimum temperature of 60°C, while cold products such as deli meats and cold cuts should be stored and served between 2 and 5°C. Under most conditions, the minimum serving temperature of 60°C for warm samples should be adhered to as the maximum volatile aromatics will be detected at this temperature. However, for consumer studies on whole-muscle meats designed with endpoint temperatures to meet the consumer preferences, lower serving temperatures may be needed. Under these conditions, it is recommended that meat be held at 49°C for no more than 20 min in a glass, covered container so that the sensory properties of the steaks and roasts are not affected. If samples are at 49°C when cut, placed in the serving container, and served, the sample should be about 38°C. To humans, the sample will be warm but not cold. Samples should be held as a whole sample and portioned prior to serving to minimize sensory changes due to holding.

When cooking larger cuts of whole-muscle meats to constant temperature endpoints, appropriate time spacing between samples may be difficult to achieve due to different rates of cooking. Some samples may need to be held in a warm environment. Panelists should be informed so that they understand that time between samples (a minimum time should be defined and adhered to) may vary and be longer than the minimum.

Ideally, the samples will be served immediately after being prepared. For fresh meats, this includes serving the samples immediately after being cut, with panelists receiving cubes from various locations within the steak or, in the case of larger portions for consumer panelists, the same steak portion for all treatments. If samples need to be held prior to serving, such as for foodservice types of products, preliminary tests need to be conducted to ensure that holding methods do not affect color, flavor, or texture of the samples. Several procedures for maintaining the temperature of the samples have been used, which include the use of the following sections.

Serving containers and utensils

Serving containers and utensils must be made of inert materials that are odor free and nonreactive so that they do not impart any odors or flavors to the samples (ASTM Standard E1871-17, 2017). Serving containers must also be able to maintain sample temperature and moisture content and should be neutral in color unless tint is needed to mask color differences. Glass or white-glazed china (white-glazed porcelain) that has been washed in an unscented detergent, followed by baking at 93°C for several hours, is best but may not be practical for large studies. In these cases, plastic containers and utensils can be used. It should be noted, however, that some plastic materials are less inert, more susceptible to temperature changes, and less odor free than others. All plastic containers should be pretested prior to their use to ensure that under the conditions of use they do not impart sensory changes to the product. The use of wooden toothpicks when serving cubed samples could transfer flavors to the meat and disrupt textural integrity of the sample and should be avoided when possible.

It is recommended that glass containers, such as glass custard dishes, plates, bowls, and other glass containers with lids or concave watch glass covers be used, especially when aroma compounds are to be evaluated. The lids will also keep the samples from drying out while being held. If lids contain material other than glass, preliminary tests should be conducted to make sure the lids do not impact the sensory properties of the samples.

Samples should be coded with 3-digit random numbers. Materials used to mark containers with 3-digit random numbers should be odor free, such as grease pencils, low-odor markers, or labels with printed numbers. If Sharpie^® pens are used, the codes should be placed on the containers 24 h prior to use, and the container should sit uncovered to air for 24 h prior to use. Avoid an alphabetical or numerical order to minimize bias.

Number of samples per session and order of presentation

The number of samples that should be presented in a given session is a function of the following: test method—discrimination panel vs. descriptive panel vs. consumer; product characteristics—flavor intensity and spiciness; level of panelist expertise and/or experience; sensory and mental fatigue; number of attributes and modalities (flavor and texture vs. just texture only) to be measured per sample; and use of palate cleansers.

Sensory panel participants’ informed consent

The Belmont Report entitled “Ethical principles and guidelines for the protection of human subjects of research” was created by the Department of Health, Education, and Welfare in 1979 to safeguard human subjects used for research, including sensory panels (Department of Health, Education, and Welfare, 1979). This report protects human subjects used in research by fulfilling 3 fundamental ethical principles of respect for persons, beneficence, and justice. These principles protect the autonomy of participants, ensure that they are treated with courtesy and respect, and provide for informed consent. The report maximizes benefits for the research project while minimizing risks to the research subjects and ensures that reasonable, nonexploitative, and well-considered procedures are administered fairly.

In the United States, federally funded research and research conducted at state and federally supported institutions involving human subjects such as sensory panels must obey the ethical rules of the report. Therefore, participants’ informed consent must be obtained, and supervision of the research must be conducted by an institutional review board (IRB) unless the procedures are exempt due to the service nature of the work and not publishing research results. The form prepared for obtaining participants’ consent should precisely convey all the information about the sensory panel to the participants. The consent form should include the purpose of study and study design, who can participate, who will be conducting the research, what participants will be asked to do, possible risks and discomforts, possible benefits/compensation, and who to see with problems or concerns. Figure 16 shows an example of a sensory panel informed-consent form that must be adjusted specifically for each study. For research, the IRB number of the approved procedures should be included in the manuscript.

Figure 16.

Example of an informed-consent form document (Department of Health, Education, and Welfare, 1979).

Discrimination methods

Discrimination or difference testing methods are used to determine when there is a detectable difference between samples. Trained or untrained panelists can be used, depending on the test objectives. If consumers or untrained panelists are used, it is understood that consumers are not as sensitive to differences between products. When trained panelists are used, smaller differences may be discernible.

Discrimination testing methods are divided into 2 broad categories: (1) testing for overall difference and (2) testing for differences in specific attributes. Sensory professionals select the test based on the test objectives. A brief description of the 5 most commonly used discrimination tests used in meat research is included below. Other methods are used and discussed in Meilgaard et al. (2025), Stone and Sidel (2004), and Lawless and Heymann (2010).

The number of panelists required for a given test should be determined prior to conducting the test, regardless the test method used. The recommended number of panelists is based on the test objectives (similarity vs. difference testing), the desired level of test sensitivity, and the amount of variability within samples. Test sensitivity is a function of the levels of α and β, and the levels of either P_d (triangle and duo-trio tests) or δ (duo-trio and tetrad tests). α is the probability of committing a Type I error, which concludes that a perceivable difference exists when, in reality, there is no difference. β is the probability of committing a Type II error, which concludes that no perceivable difference exists when, in reality, there is a perceivable difference. The proportion of discriminators, or P_d, is the proportion of the population represented by the assessors (panelists) that can discriminate between the 2 products in the test. The δ is the effect of size or maximum allowable sensory difference to be detected in the test.

The following guidelines can be used in setting the levels for these factors (ASTM Standard E1885-18, 2018; ASTM Standard E2610-18, 2018; ASTM Standard E3009-24, 2024; Meilgaard et al., 2025). For α, a statistically significant result at 10 to 5% (0.10–0.05) indicates “slight” evidence that a difference is apparent, 5 to 1% (0.05–0.01) indicates “moderate” evidence, 1 to 0.1% (0.01–0.001) indicates “strong” evidence, and less than 0.1% (<0.001) indicates “very strong” evidence. For β, the strength of evidence that a difference is apparent is assessed using the same criteria as above but substituting “is not apparent” for “is apparent.”

When conducting triangle and duo-trio tests, P_d, the maximum allowable proportion of discriminators, falls into 3 categories of P_d being less than 25%, representing small values; P_d being greater than 25% but less than 35%, which represents medium-sized values; and P_d being greater than 35%, representing large values. For duo-trio and tetrad tests, δ, the maximum allowable sensory difference, falls into 3 categories where δ is less than 0.5, representing small values; δ is greater than 0.5 but less than 1, representing medium-sized values; and δ is greater than 1.0, which represents large values.

The actual settings for α and β should be dictated by the level of risk of being wrong in the outcome of the test. In food companies, these risk settings are influenced by the tier level or volume of the brand or product, the degree of product change, and the known sensory variability of the product. α, β, and P_d or δ may be adjusted based on these criteria and historical company practices. Consult a statistician and/or ASTM Standard E1885-18 (2018), ASTM Standard E2610-18 (2018), or ASTM Standard E3009-24 (2024) for recommendations.

Triangle tests

The triangle test is an overall difference test and is believed to be very sensitive. The test does not, however, give directional or attribute information. There is a guessing rate of 33% that is accounted for in the statistical tables, used to determine if a difference exists or not. To conduct the test, panelists are presented with samples either one at a time or all at once. The panelist is asked to evaluate each sample in the specified order and to identify the sample that is different. The instructions define that 2 samples are the same and 1 is different (Figure 17). Each sample should be presented with a 3-digit code, and order should be randomized for each panelist. There are 6 potential serving orders where A and B represent the 2 products: AAB, ABA, BAA, ABB, BAB, and BBA.

Figure 17.

Sample ballot for triangle testing. Adapted from Meilgaard et al. (2025).

Table 19.7. from Meilgaard et al. (2016) can be used to determine how many panelists are needed based on the predefined levels for α, β, and P_d. If the study objective is to determine whether a detectable difference exists between treatments, the value selected for α is typically smaller than the value selected for β. For example, with α equal to 0.05, β equal to 0.20, and P_d equal to 30%, 40 panelists would be required. If the study objective is to determine if samples are sufficiently similar to be used interchangeably, however, the value selected for β is typically smaller than the value selected for α. For example, with an α equal to 0.20, β equal to 0.05, and P_d equal to 30%, 39 panelists would be required. Results are analyzed by comparing the number of correct responses to the total number presented using the predetermined α, such as in Table 19.8 in Meilgaard et al. (2016). Additional information on how to conduct this test can also be found in ASTM Standard E1885-18 (2018).

Duo-trio tests

The duo-trio test is an overall difference test and is easier for consumers to understand than the triangle test. There is a 50% guessing rate with the duo-trio test, so it is statistically less efficient than the triangle test. Researchers, therefore, need more respondents to conduct this test than the triangle test. Panelists are served a control. After evaluation of the control, panelists are served 2 samples: one is the same as the control and the other is different. They are asked to identify the sample that is the same as the control (Figure 18). Again, directional and attribute information is not obtained. The levels of α, β, and P_d for the test are predetermined based on whether one is testing for difference (α is set smaller than β) or similarity (α is set larger than β). This determines the number of respondents needed as shown in Table 19.9 of Meilgaard et al. (2016). The recommended number of panelists is 69, with α equal to 0.05, β equal to 0.20, and P_d equal to 30% (testing for difference), and 68, with α equal 0.20, β equal 0.05, and P_d equal to 30% (testing for similarity). The control sample rotates between the 2 samples and the order of presentation of the 2 samples rotates as well. Based on this, the serving order with the first sample defined as the control is ABA, AAB, BAB, and BBA. After completion of the test, the total number of responses and the number of correct responses are counted. Using the defined α, the number of correct responses needed to determine a difference is identified in Table 19.10 of Meilgaard et al. (2016). Additional information on how to conduct this test can also be found in ASTM Standard E2610-18 (2018).

Figure 18.

Sample ballot for duo-trio testing. Adapted from Meilgaard et al. (2025).

Degree of difference

This method also know as Difference-from-Control can be used to determine not only if a detectable difference exists but also the size of differences among a set of samples. Difference-from-control (DFC) testing is also recommended when products are highly variable. This method can be expanded to measure not only the degree of overall difference but also differences in multiple attributes, such as beef-flavor strength or tenderness.

In DFC testing, panelists receive the control sample twice, one time as a labeled reference sample and a second time served blind as a coded sample along with the other coded test samples. Panelists are instructed to taste the reference first, then each coded sample, and indicate the degree of difference from the reference sample using a scale from “none” to “very large difference,” as shown in Figure 19. The scales are typically 7 to 10 points and can be line scales or category scales that are fully anchored or anchored only at the ends (0 = no difference and 7 = very large difference). Twenty to 50 panelists are generally recommended (Meilgaard et al., 2016). The number of panelists required is determined by the amount of variability within samples and the test objective. If the test objective is to detect a difference, the emphasis will be on Type I error. If the objective is to test for similarity, the emphasis will be on Type II error. Testing for similarity will require more sample presentations (panelists) than testing for difference.

Figure 19.

Sample ballot for difference-from-control testing. Adapted from Meilgaard et al. (2025).

This method can also easily accommodate multiple treatments in a study. With multiple treatments, it is best to present all samples at the same time with clear instructions on the order in which samples are to be evaluated. If, however, it is not possible to present all samples at the same time due to issues such as taste bud fatigue, present only one test sample at a time along with the reference.

Results are analyzed by calculating the mean difference between the identified control and each test sample vs. the difference between the identified control and the blind control. The mean differences are then compared using ANOVA if there are multiple test samples in the study or a paired t-test if there is only one test sample. Additional details on this method and data analysis procedures can be found in Meilgaard et al. (2016).

Tetrad tests

This method can be used to determine whether a perceptible sensory difference exists between 2 samples. The samples may differ in a single sensory attribute or in several attributes. In a tetrad test, samples are served all at the same time if possible. Panelists are presented with 4 coded samples, where 2 samples are from 1 product and the other 2 are from the other product being tested. The panelists are asked to taste the samples in the order indicated and group the 4 samples into 2 pairs based on the level of similarity between samples (Figure 20).

Figure 20.

Sample ballot for tetrad test. Adapted from ASTM E3009 (2024) and Meilgaard et al. (2025).

There are 6 potential serving orders, where A and B represent the 2 products: AABB, ABAB, ABBA, BBAA, BABA, and BAAB. The recommended number of panelists is based on test objective (testing for difference or similarity) and predefined α, β, and δ. Table 1.1 in ASTM Standard E3009-24 (2024) can be used when testing for difference, and Table 1.2 can be used when testing for similarity. When testing for differences, the recommended number of panelists is 30, with α equal to 0.05, β equal to 0.20, and δ equal to 1.25. The recommended number of panelists when testing for similarity is also 30, with α equal to 0.20, β equal to 0.05, and δ equal to 1.25. The results are tallied, and the level of significance determined by referring to Table A1.3 in ASTM Standard E3009-24 (2024) if testing for a difference and Table A1.4 if testing for similarity.

Because of the nature of the task that the panelists must perform, this method is more efficient statistically than the triangle test or the duo-trio test. However, this method may be less suitable for samples with a high degree of variability or that may cause excessive sensory fatigue or carryover effects. Additional details on this method can be found in ASTM Standard E3009-24 (2024).

Descriptive analysis methods

Descriptive sensory analysis involves sensory approaches that discriminate and describe both qualitative and quantitative properties using trained panelists. Considerable time and effort are involved in training and maintaining descriptive sensory panels. All sensory attributes of a meat product can be subjected to descriptive analysis or limited to selected criteria such as flavor and texture. The 4 prevalent methods are described below. For a more complete discussion and understanding of descriptive analysis methods, see ASTM MNL13-2^ND-EB (2020).

Texture-profile method

The texture-profile method is based on the same concept as flavor profiling in that the overall texture of a product comprises a number of different texture attributes. Panelists define the procedures and terminology to use in the textural evaluation. Modifications are made in the texture-profile approach as outlined in ASTM MNL13-2^ND-EB (2020) compared to the flavor-profile method. For the texture-profile method, more precise definitions and evaluation procedures were developed. New reference scales and various scaling procedures were also defined. With the texture-profile method, there is a collection of individual scores without consensus and statistical analysis of data.

Texture attributes in the texture-profile method can be classified as either mechanical, geometric, or those related to moisture and fat content. Mechanical characteristics are revealed as the meat samples react to stress, such as chewing. Geometric characteristics relate to size, shape, and orientation of the product before and during breakdown. The contributions of moisture and fat are determined through mouthfeel. Texture references have been developed and can be used for training. An example of a texture lexicon for ground beef patties was published by AMSA (1983). A lexicon is defined as a group of attributes, their definition, and references within a product or product type. An example ballot that includes flavor and texture attributes for ground beef patties is presented in Figure 21. This ballot uses beef-flavor attributes from the beef-flavor lexicon (Adhikari et al., 2011), as discussed below, and texture attributes from AMSA (1983). The texture attributes can be easily defined and scaled using solid oral texture attributes from Meilgaard et al. (2016). Note that not all texture attributes for ground beef were included on the ballot—only those important to the hypothesis of the study were included.

Figure 21.

Example of a flavor and texture descriptive attribute ballot for ground beef patties based on AMSA (1983), Meilgaard et al. (2016), and Miller et al. (2019); panelists provide ratings for each attribute from each sample provided.

Quantitative-descriptive analysis

Quantitative-descriptive analysis (QDA) was developed to provide a stronger statistical treatment of data than data from profile-type methods. While training is provided regarding methods and terminology, to some degree panelists are free to develop their own approach to scoring. Overall discussions of the evaluations after a session are not usually conducted. Data are reported in the form of a spiderweb with a branch or spoke for each attribute. An example of this form of data reporting is shown in Figure 22. Additional information on this method can be found in ASTM MNL13-2^ND-EB (2020).

Figure 22.

Spider web from quantitative-descriptive analysis wherein 3 levels of sodium lactate were added to pork roasts.

Meat-descriptive attribute evaluation

In the first guidelines in 1978, a unified method of evaluating meat palatability attributes was introduced (AMSA, 1978). This method concentrated on juiciness, connective tissue amount, muscle-fiber tenderness, and overall tenderness as the major attributes of meat palatability. The 2 previous versions of the guidelines also included overall flavor intensity. This attribute is a general intensity attribute and does not describe a specific flavor attribute. For example, 2 samples may have the same intensity of overall flavor; however, one sample may have predominantly beef identity and browned flavor aromatics, and the other sample may have predominantly liver-like and putrid flavor aromatics. Also, it is difficult to reference and scale across panels. When possible, it should be replaced with the species-specific flavor lexicons, where each flavor attribute can be referenced and scaled. If it is not practical to train for the entire flavor lexicon or if it is not necessary based on the experimental objectives, 1 or 2 flavor notes, such as beef-flavor identity and fat-like flavor aromatics, could be used (Munoz et al., 1992; Meilgaard et al., 2025). Selection of attributes to include in the lexicon should be based on the study objectives. This method uses an 8- or 9-point, verbal anchored scale (Figure 23), or it can be modified to a 15- or 16-point scale to more easily incorporate flavor attributes (Figure 24). Regardless of scale, the method requires training on each attribute, panel performance validation, and use of correct cooking and product sampling techniques.

Figure 23.

Example ballot for beef descriptive flavor and texture attributes using 15- and 16-point scales (Miller, 2013).

Figure 24.

Examples of training sessions for a beef-flavor descriptive attribute panel using the beef-flavor lexicon (Adhikari et al., 2011; Miller et al., 2023).

Selection and training sensory panelists for discriminative or descriptive testing

Before initiating panelist training, the researcher must determine which sensory test method is most appropriate to fulfill the objectives of the study. Various test methods are available, and it is not the intent of these guidelines to fully describe these procedures. It is suggested that ASTM MNL26-3^RD-EB (2020) and Meilgaard et al. (2025) be reviewed to determine the most appropriate procedure. A short discussion of the most common methods used was presented in the previous sections. Panelists can be trained to participate in discriminative and descriptive testing.

Prescreening candidates

Potential candidates are first prescreened to determine their level of interest, availability, dependability, health (including dentures, allergies, use of medication), work experience, gender, age, smoking/tobacco use status, and food likes/dislikes (Figure 25). Background information on prospective panelists is valuable in selecting those individuals who have the greatest potential to become effective panelists.

Figure 25.

Example of prescreening questionnaire for selection of panelists for a meat texture and flavor descriptive attribute sensory panel (modified from Meilgaard et al., 2016).

During the prescreening process, a candidate’s ability to follow directions or make concise judgments should also be determined because panelists who are not able to do these tasks will not be successful panelists. Logic tests can be conducted to determine a candidate’s ability to follow directions and make decisions. Figure 26 provides an example of logic tests, and additional logic tests can be found in Meilgaard et al. (2016). Candidates should mark the line in the approximate area for a correct response. Those who provide incorrect answers by marking the line in the opposite direction as requested should be immediately eliminated. These candidates will not follow instructions, and the panel leader will have to continually work with them to ensure that they understand. The result can be a decrease in motivation and positive attitude of other panelists. When grading, a 70% or above is considered acceptable on a logic test.

Figure 26.

Examples of logic tests used for screening potential sensory panelists. Adapted from Meilgaard et al. (2016).

Screening candidates

Candidates who have passed the prescreening stage are invited to participate in a series of screening exercises. The purpose of screening is to identify candidates who have the required sensory and cognitive skills to perform the descriptive analysis tasks (ASTM MNL90-2^ND-EB, 2023). Candidates must therefore be screened for the ability to (1) detect differences in product characteristics and in their intensities, (2) describe those characteristics using verbal descriptors and scaling methods for the different intensity levels, and (3) perform abstract reasoning as descriptive analysis relies heavily on the use of references when characteristics must be quickly recalled and applied to other products (Meilgaard et al., 2025).

It is often necessary to screen 4 times more panelists than what will be needed in the actual panel. Sensory protocols and procedures during screening should be similar to those used later in actual studies. It is important to use the same meat products for screening that will be evaluated by the panel after training. A higher level of discrimination is required for descriptive analysis panelists than for discriminative tests. The screening procedures should be more rigorous for descriptive analysis panelists and will require that more panelists be screened to achieve the final number of panelists needed. Screening panelists can include more than one of the strategies discussed below.

Matching tests are designed to determine a potential panelist’s ability to describe or identify descriptive attributes. Meat products contain multiple flavors and textures, especially further-processed products. Even if the original objective of the trained panel is to evaluate whole-muscle beef steaks, panelists need to have the ability to describe, identify, and rate flavor attributes. If the panelists can describe texture attributes of a further-processed meat product, such as a frankfurter or sausage product, the panel can be expanded or can be further trained to evaluate these products. Meilgaard et al. (2025) described these tests in detail. It is useful to present matching tests for taste and aroma to better assess a candidate’s ability to discriminate. Table 14 provides examples of matching tests from Meilgaard et al. (2025). The candidate is served a set of cups labeled sweet, sour, salty, bitter, and water, as well as a set of 5 cups coded with 3-digit random numbers. The labeled cups contain the appropriate stimulus at just more than threshold levels, and the coded cups contain a corresponding stimulus. The candidate is instructed to taste the labeled cups first to familiarize themselves with the basic taste. They are then asked to taste the coded samples and indicate on a score sheet the matching code number for each stimulus. Panelists should use palate cleansers between evaluating each sample.

Table 14.

Suggested samples for a matching taste test, adapted from Meilgaard et al. (2025)

Basic Taste	Stimulus	Concentration (g/L)
Sweet	Sucrose	20
Salty	Sodium chloride	2.0
Sour	Citric acid	0.5
Bitter	Caffeine	0.6
Astringent	Alum	0.1%
Water	Filtered water	—

Sniff tests are conducted for matching tests for aromas (Meilgaard et al., 2025). To conduct sniff tests, cut 0.75-cm-wide strips of Whatman filter paper at least 2.54-cm long. Cotton balls can be used as an alternative to Whatman filter paper. Obtain essential oils representing a variety of aromas from a business that uses essential oils to formulate food or personal care products (Table 15). Dip the filter paper strip in the oil sufficiently to concentrate the oil’s aroma. Allow the strip to dry under a hood for 30 min, then place it in a glass jar with a lid, a glass sniffer with a glass concave watch top, or in a test tube with a lid. Label the containers with random 3-digit codes. Ask the candidate to remove the lid or push the lid aside and sniff the contents of the container without touching the contents. Panelists can shake the container with the lid intact to concentrate the aromatics before evaluation. Candidates are asked to match the aroma they detect with the appropriate aroma descriptor selected from a list at the bottom of the score sheet; list 2 to 4 more descriptors than the number of stimuli presented. Candidates should be accepted if they are able to correctly identify approximately 70% of the attributes that are presented.

Table 15.

Suggested samples for sniff tests, adapted from Meilgaard et al. (2025)

Aroma Descriptors	Stimulus
Anise, licorice	Anise oil
Almond, cherry	Amaretto, benzaldehyde, oil of bitter almond
Cinnamon	Cinnamaldehyde, cassia oil
Clove, dentist’s office	Eugenol, oil of clove
Ginger	Ginger oil
Green, freshly mown lawn	cis-3-Hexenol
Lemon, lime, or orange	Lemon, lime, or orange oil
Peppermint, minty	Peppermint oil
Vanilla	Vanilla extract
Wintergreen	Methyl salicylate, oil of wintergreen

Ranking tests are used to determine a candidate’s ability to discriminate graded levels of intensity of a given attribute. As described in Meilgaard et al. (2025), candidates are presented with a series of samples in random order. The samples are coded with 3-digit random numbers and cover a range of a specific attribute. Ask the candidates to rank the samples in order of increasing intensity of the stated attribute. Examples of sample sets are listed in Table 16.

Table 16.

Suggested sample sets for ranking tests, adapted from Meilgaard et al. (2025)

Attribute	Sensory Stimuli	Concentration (g/L)
Taste
Sweet	Sucrose/water	10	20	50	100
Salty	Sodium chloride	1	2	5	10
Texture
Hardness	Cream cheese,a American cheese,a peanuts, carrotb
Juiciness	Banana,b carrot,b mushroom,b appleb

• 1.27-cm cubes.

• 1.27-cm slices.

Sniff tests can also be used as identification tests to understand a panelist’s ability to recognize, describe, or identify descriptors. These tests are mainly used in the selection of descriptive panelists. Sniff tests, as described above, can be used. Jars with varying flavor aromatics are presented individually to panelists using 3-digit random codes. The candidate is asked to describe the attribute that they detect on a sheet of paper. When evaluating their score sheet, a correct answer is when the candidate uses descriptors similar to the attribute. For example, if cedar oil is used in the sniff test, correct answers could be cedar, wood, a forest, sweaters, stored sweaters, or a wooden chest.

Triangle tests are recommended to determine a panelist’s ability to discriminate differences in sensory attributes. Duo-Trio tests can be used for this exercise, but are not as sensitive, so more duo-trio tests are required to have the same sensitivity as triangle tests. For discriminative panel selection, fewer triangle tests need to be run (from 6 to 8); for descriptive panelists, a larger number of triangle tests should be conducted to provide a greater range of attributes (10 to 15).

A sequential analysis procedure is used to minimize the number of tests needed for screening (Bradley, 1953). That procedure facilitates an early decision on the suitability of panelists to participate as panel members. The sequential procedure makes one of the following decisions after each triangle test: 1) Accept the candidate as a potential panelist; 2) reject the candidate; or 3) continue testing. The decisions are based on specifications of 4 parameters: P₀ = Maximum proportion of correct decisions ruled as an unacceptable candidate; P₁ = Minimum proportion of correct decisions ruled as an acceptable candidate; α = Probability of selecting an unacceptable candidate; and β = Probability of rejecting an acceptable candidate. By plotting test numbers against the accumulated number of correct test results, a decision is made based on the region in which the point is plotted (Figure 27). The regional boundaries are described in greater detail by Cross et al. (1978) and Meilgaard et al. (2025).

Figure 27.

Sequential analysis chart used for screening potential panelists. Adapted from Cross et al. (1978).

Examples of triangle tests used in the screening procedure are shown in Table 17. Test samples for triangle tests are prepared to give a 2-unit difference on an 8-point scale, or a 3- to 5-point difference on a 16-point scale for the attribute being tested. In this example, the attributes are tenderness, juiciness, and connective tissue amount. More than 5 triangle tests are recommended, but these 5 examples can be used as an example of how to create differences. Differences should be of a magnitude that could easily be detected by an experienced panel leader. The values P₀ = 0.45, P₁ = 0.70, α = 0.10, and β = 0.10 are used (Meilgaard et al., 2025). Figure 27 is then applied to accept, continue testing, or accept panelists.

Table 17.

Examples of triangle tests that can be used and the sensory attribute of interest

Triangle Tests	Explanation of Attribute Differences Targeted for Testing
1.	Strip loin steak cooked to 70°C and a strip loin steak cooked to 75°C, with juices pressed out to get differences in juiciness.
2.	Eye of round steak, cooked to 70°C, and a strip loin steak cooked to 70°C, to get differences in tenderness.
3.	Top sirloin steak and a strip loin steak, both cooked to 70°C to get differences in flavor intensity (or ST muscle “leached” in water for a few hours before cooking).
4.	Ground beef patty and a ground beef patty containing 1.0% added ground liver, evenly mixed and distributed to get differences in liver flavor.
5.	Brisket or bottom round steak grilled, and a strip loin steak grilled to get differences in connective tissue.
6.	Chicken breast marinated with salt and phosphate, and chicken marinated with salt and a natural flavor phosphate replacer.
7.	Pork sausage made from prerigor pork and postrigor pork.

• Abbreviations: ST, Semitendinosus.

At the end of the screening period, the candidates in the “accept” region can be selected for training. If time is a factor, screening can be stopped after 10 sessions, but when evaluating for multiple attributes, a higher number of tests provides greater insight into the panelist’s ability to discriminate. It is desirable to select only those in the “accept” region, but if most of the candidates are in the “continue testing” region, they also could be selected. Another option would be to recruit more candidates and reinitiate the screening process. In panel selection, screening should not be considered a part of training, but rather a test to quickly eliminate those individuals who cannot detect large differences in attributes. At least twice as many individuals should be screened as are needed on the final panel.

Meilgaard et al. (2016) suggest that, when using triangle tests, to reject candidates scoring less than 60% correct on the easy tests or less than 40% on the moderately difficult tests. When using duo-trio tests, reject candidates scoring less than 75% on the easy tests or less than 60% on the moderately difficult tests. After completion of the screening tests, accept panelists for training who pass all stages (ASTM MNL90-2^ND-EB, 2023). During candidate testing, observe panelist behavior, timeliness, ability to interact with others, confidence level, and dependability. Do not accept a panelist who misses an appointment, is difficult to work with, or is disruptive or domineering in a group setting. Panelists who are not dependable during a “job interview” will not be dependable after they have the job. If the panel leader determines that an individual is disruptive, the panelist’s behavior pattern will be difficult to alter and might take more effort than desired to change. Final panelists selected for training should be acceptable based on all criteria discussed above.

Consumer sensory panels

While the use of a trained panel to assess treatment differences in a set of products is critical in determining if the treatment differences are detectable, it is also critical to know if these differences impact consumer acceptance. Are these differences large enough for consumers to detect? Are they important enough to affect overall acceptance? Consumer testing is often a key part of a study, as the fate of a food product rests on acceptance by the consuming public. It is only through careful planning of the consumer test, including choosing the most appropriate test method, that the researcher can get reliable, repeatable results that can be projected to the real consumer population.

Trained and consumer panels can both be useful to determine if differences exist within and between products. Consumer tests should be used if it is important to determine if consumers can detect the differences in the product due to the addition of some treatment. In this situation, Triangle or Tetrad tests can be conducted with consumers to determine if the consumers detect a difference between the products. An example of this type of test would be the addition of a microbiological intervention to ground beef and its impact on the sensory properties of the product. As consumers are less likely to detect differences compared with trained panelists, it is advisable to use consumer Triangle or Tetrad test data in conjunction with trained panel data when making decisions on whether or not there is a perceivable difference between the products.

Additional points to consider when designing and conducting a consumer sensory test are experimental design, protocol design, test method, ballot development, moderator selection, and interaction with the sensory subjects. Meilgaard et al. (2016, 2025), Lawless and Heymann (2010), ASTM MNL26-3^RD-EB (2020), and Stone and Sidel (2004) are excellent references for information on the design of consumer panels.

Ballot development

The consumer sensory ballot, which is defined as the sensory instrument for consumer testing, is critical to ensure that the sensory professional is accurately testing consumer responses and not biasing the responses with a poorly designed ballot. In designing a ballot, some general rules are to ask for responses in the order in which they are normally encountered while eating or sampling the product. For example, most consumers will evaluate the appearance, then flavor, and lastly texture. A combination of close-ended (i.e., hedonic scales) and open-ended (no defined response) questions should be used. Scales should be consistent with the ballot for the number of points of determination, scaling, and style. For example, if a 9-point hedonic scale is being used, do not change to a 5- or 7-point hedonic scale across questions within the same ballot. It is appropriate, however, to include intensity (5-, 7-, or 9-point) and/or JAR (typically 5-point) scales for impressions of specific attributes. When using JAR scales, use penalty analysis to determine the impact of an attribute not being JAR. Do not reverse the end-anchors on the scales within one ballot. Provide clear, concise directions and unambiguous questions to ensure that the test accurately measures consumers’ sensory perceptions and keeps questions to a minimum. When selecting the number of points on a scale, a decision has to be made to have either an even or an odd number of points. Odd-numbered scales provide a neutral point that can be anchored or not anchored. Even point scales do not have a neutral point. In some studies, the researcher may need to understand if the consumer is neutral or does not have a defined like or dislike of the product, or it may be important to force the consumer to make a decision for like or dislike. Both types of scales are used, but the researchers should understand how the scale impacts consumers’ thinking and perception.

The most important consumer question is almost always the consumer’s overall like/dislike of the product. The position of this question can influence the answer. When asked about the appearance and taking the first bite of the product, the answer reflects the consumer’s first impression. When asked at the end of the ballot after questions concerning the appearance, juiciness, tenderness, and flavor, the answer reflects the consumer’s first impression and any influence of the questions asked. The researchers must consider this when designing and interpreting the results.

Hedonic scales

The hedonic scale can be shown horizontally or vertically, and scales can be verbally anchored at each point along the scale with 9 categories or as continuous line scales (Figure 28). When using the hedonic scale horizontally, it should be anchored with “Dislike Extremely” on the left and “Like Extremely” on the far right (Figure 28). The purpose of anchoring the scale at each point is to encourage a continuum with equal spaces between each successive increase in like/dislike or preference.

Figure 28.

Examples of commonly used hedonic and just-right scales.

Meilgaard et al. (2025) and Lawless and Heymann (2010) give examples of hedonic scales that can be used in consumer testing. These scales range from the anchored, 9-point hedonic, to the end-anchored, 9-point hedonic; to the end- and neutral-anchored, 9-point hedonic; to the non-balanced scale (more or less categories of like in relation to dislike). Many different forms of hedonic scales can be used without major effects on the value of the results, as long as the essential feature of verbal anchoring of clearly successive categories is retained. There should be at least 5 categories. Replacement of the verbal categories with caricatures representing degree of pleasure and displeasure (smiley scale) has been used, but studies on this type of scale have indicated potential issues with how children interpret the caricatures, so these scale types are not recommended (ASTM Standard E2299-13, 2021). If testing with children, a 9-point “super good/super bad” scale developed by Kroll (1990) has been used successfully. It can be truncated to 7 points for younger children if needed.

Just-about-right and intensity scales

Intensity scales and JAR scales can be used to determine if products differ significantly in the levels of specific attributes. These scales can be bipolar or unipolar. Intensity scales can be line scales or category scales, and they can be end-anchored or fully anchored. When used as category scales, intensity scales are often 5-point, 7-point, or 9-point. If using a mixture of JAR and intensity scales on the ballot, it is good practice to use different scale lengths for the 2 scale types to minimize confusion about what is being asked.

JAR scales are typically category scales and are most often 5-point, with 7-point and 9-point scales used less often (Figure 28). Because an attribute can vary without negatively impacting consumer liking, it is critical when using these scales that penalty analysis be used to assess the impact or penalty of an attribute not being JAR. Penalty analysis links the drop in overall liking to the proportion of respondents rating an attribute either too high or too low. It, therefore, provides the ability to prioritize product optimization opportunities by determining the critical product attributes that penalize product acceptance the most. It is important to note that penalty analysis should only be run for JAR attributes with 20% above or below JAR. Running the analysis with less than the 20% cutoff results in too much dependency on any given respondent. For studies with smaller sample sizes or when analyzing subgroups within a study, the cutoff should be increased to at least 25%. Refer to ASTM MNL63 (2009) for additional information on penalty analysis.

Several SAS packages, such as XLSTAT, offer an automated penalty analysis feature. If unavailable, however, the penalty for an attribute not being in the JAR is calculated in the following manner:

• •

  Collapse JAR scales to 3 points.

  • o

    Too little (TL)

  • o

    JAR

  • o

    Too much (TM)

• •

  For each attribute with 20% or more of respondents rating it either too high or too low, calculate the mean drop in overall liking for not being JAR.

  • o

    Mean drop_TL = Average_JAR – Average_TL

  • o

    Mean drop_TM = Average_JAR – Average_TM

• •

  Calculate the total penalty for each attribute not JAR.

  • o

    Total penalty = (Mean drop) * (% not JAR)

Penalties can be shown graphically in 2 ways, as shown below in Figures 29 and 30. Total penalties can be categorized into different tiers based on potential opportunity to improve overall liking if the offending attribute is brought to JAR: < 0.25 = low-tier penalty: no changes needed; 0.25 to < 0.50 = mid-tier penalty: consider changing; and ≥ 0.50 = high-tier penalty: must change.

Figure 29.

Mean drop in overall liking as a function of the percent not at just-about-right ratings for the various attributes. Abbreviation: JAR, just about right.

Figure 30.

Total penalties of the various attributes.

In the above example, results showed that 35% of the panelists rated the sample “too salty,” but the drop in mean overall liking was 0.5 units, which resulted in a total penalty of 0.18 units (Figure 29). This penalty is considered low tier, indicating the salt level does not need to be adjusted. Additional information on penalty analysis and the uses and abuses of JAR scales can be found in ASTM MNL63 (2009).

An example of a consumer ballot is presented in Figure 31 using a combination of hedonic, intensity, and JAR scales. Note that demographic and product use questions can be included on a consumer ballot. Computerized demographic and use questions are used on prescreening questionnaires so that these questions do not influence consumers’ expected or actual knowledge of the project objectives (defined as expectation error). Instructions are also included on the ballot, even though a moderator should be present during testing to assist participants in understanding what is expected of them.

Figure 31.

An example of a consumer sensory ballot using 9-point hedonic scales and 5-point just-about-right scales.

Warner-Bratzler shear force (WBSF)

This measure can be obtained either with the original WBSF machine or with WBSF blade attachments to an automated testing machine (e.g., Instron, United, Texture Technologies, etc.). In addition to peak load (maximum shear force), other traits that might be useful can also be obtained with an automated testing machine. V-notch blades used for shear force should either be the blades made for WBS machines by G-R Manufacturing (Manhattan, KS, USA) or blades sold by the testing machine manufacturer (Figure 32). These blades are milled to exact specifications, including the bevel on the cutting edge. Unless in-house manufactured blades meet these exact specifications, they should not be used.

Figure 32.

Examples of acceptable cores and the Warner-Bratzler shear force blade when conducting Warner-Bratzler shear force determinations.

Warner-Bratzler shear blade specifications include: (1) blade thickness of 1.1684 mm (0.046 inches); (2) V-notched (60° angle) cutting blade; (3) cutting edge beveled to a half-round; (4) corner of V rounded to a quarter-round of a 2.363-mm diameter circle; (5) spacers providing a gap for the cutting blade to slide through of 2.0828-mm thickness. After cooking and recording the final cooked temperature and weight, steaks, chops, and filets should be chilled overnight at 2 to 5°C before coring. Chilling firms the steak, chop, and filet and makes it easier to obtain uniform diameter cores or pieces. If chilling is not used, some protocol to obtain a consistent steak temperature before coring should be followed, such as allowing steaks to reach room temperature (23°C). Round cores should be uniformly 1.27 cm (0.5 inches) in diameter and removed parallel to the longitudinal orientation of the muscle fibers so that the shearing action is perpendicular to the longitudinal orientation of the muscle fibers (Figure 32).

Cores can be obtained using a handheld coring device (cork borer) or an automated coring device (drill press with cork borer attached). Coring devices must be in good condition and sharp, or the core diameters will not be consistent and will result in spurious increased variation in shear values. Hand coring requires very careful attention to the amount of pressure used as the corer is turned. It is best practice for the same person to do all of the coring throughout the duration of a study. A minimum of 6 cores should be obtained from each sample (this might require 2 pork chops or 3 lamb chops). Cores that are not uniform in diameter, have obvious connective tissue defects, or otherwise would not be representative of the sample should be discarded. If cooked steaks/chops were chilled, cores should be kept refrigerated and covered to reduce moisture loss until sheared to maintain a consistent temperature. Obtaining uniform cores is the most critical variable in WBSF measurements.

For poultry, uniform cores are difficult to obtain. Lyon and Lyon (1991) used 1.9-cm X 1.9-cm strip sections that run parallel to the fiber direction in the breast for shear determination. A minimum of 2 strips per filet are sheared perpendicularly to the fiber direction. Another sampling method, referred to as a strip technique (Schilling et al., 2003; Figure 33), is used by some labs. This technique uses paired scalpels that are bolted together so that the cutting edge is exactly 1.0-cm apart (G-R Manufacturing, Manhattan, KS). The paired scalpels are used to make a cut parallel to the muscle fibers. The paired scalpels, in essence, act as a subsequent cutting guide. Typically, 2 to 3 cuts are made with the scalpel using the previous cut as the guide for the next cut. After the initial cut, a sharp single-blade meat-cutting knife is used to complete the cut. The 1-cm thick strip is then tipped on its side to identify the muscle-fiber orientation for the next cut. Once the final cut is made, the width and thickness of the strip should be measured to confirm the 1-cm dimension has been achieved. One advantage of the strip technique is that many more samples for shearing can be obtained in comparison to the coring method. This particular method is very useful for obtaining samples from cooked broiler breasts.

Figure 33.

Strip technique for poultry using 190 × 190 mm strip. Picture is an example of a Warner-Bratzler setup.

All values obtained should be used for mean calculation, unless visual observation indicates a valid reason a value should be discarded (e.g., a piece of connective tissue in the shear plane). Each core should be sheared once in the center to avoid the hardening that occurs toward the outside cooked edge of the sample. WBSF tests using automated testing machines should be conducted with a crosshead speed of 200 to 250 mm/min. Shear tests that do not follow these equipment or sample specifications should not be referred to as WBSF (such as square holes in the shear blade, square meat samples, straight-edged shear blade, or blade not properly beveled, etc.). For a demonstration of WBSF, watch the video at http://www.meatscience.org/sensory.

For tenderness of whole-muscle cuts, a critical problem has been the diversity in shear force measurement protocols that has made it nearly impossible to appropriately compare shear force values among data published by different institutions. The large variability in mean shear force and repeatability of shear force from different institutions and different protocols has been demonstrated by comparing WBSF and slice shear force (SSF) values on matched steaks from the same animals (Wheeler et al., 1997b; Wheeler et al., 2007). Those data highlight not only the importance of standardized protocols, but also limitations associated with applying tenderness thresholds, such as those reported by Shackelford et al. (1991) and Martinez et al. (2023), which relate WBSF values to trained sensory panel tenderness ratings. Because WBSF and SSF measurements differ substantially and are sensitive to protocol-specific factors, such thresholds may be valid only for data generated using the same protocol, within the same institution, or within a single study.

It is often useful to be able to compare objective measures among institutions, but also make useful comparisons of consumer tenderness/texture data with objective measures of tenderness/texture. Toward this goal, much progress has been made on the use of more standardized protocols. When there is not one methodology that is superior to others, the advantages of everyone having comparable data outweigh the disadvantages of a restrictive protocol. A superior method, however, should not be discarded for the sake of uniformity of methodology. It is also recognized that the recommendation of certain procedures in these guidelines should not discourage research to evaluate alternative methods or instruments.

Numerous devices have been tested for their ability to measure meat tenderness. The measurement most often used has been WBSF, which is still frequently used today. Beef longissimus WBSF of 1.27-cm diameter cores are highly repeatable when measurement protocols are executed properly (Wheeler et al., 1994. 1996, 1997a). Several sources of error have been identified that contribute to errors in shear force assessment within and among institutions (Wheeler et al., 1994, 1996, 1997a). Available data have not demonstrated that more accurate or repeatable data are obtained when WBSF is conducted on chilled samples compared to room-temperature samples (Wheeler et al., 1994). Thus, although empirical observation indicates it is easier to obtain more uniform diameter cores from chilled steaks than room-temperature steaks, steak temperature at coring may not be critical as long as it is consistent within an experiment.

Slice shear force

Shackelford et al. (1999a) developed a simplified technique for measuring beef longissimus shear force, which is referred to as SSF. The repeatability of SSF (0.89) exceeded repeatability estimates (0.53 to 0.86) that have been reported for longissimus WBSF (Wheeler et al., 1996, 1997b), and Shackelford et al. (1999a) reported that SSF was more accurate than WBSF. Because of time constraints associated with online assessment of meat tenderness, some aspects of the SSF protocol that Shackelford et al. (1999a) developed for online assessment of beef longissimus tenderness may not be necessary or desirable for routine collection of shear force data in a laboratory setting. Thus, Shackelford et al. (1999b) conducted a series of experiments to develop an optimal protocol for routine SSF measurement in research and to evaluate SSF as an objective method of assessing beef longissimus tenderness. Subsequent experiments were conducted to adapt this technique for use on pork longissimus, lamb longissimus, and many other beef muscles (Shackelford et al., 2004a and 2004b).

Because the SSF technique allows for a substantial increase in laboratory throughput, it has allowed for the development of novel tenderness management systems. To date, by far, the most common use of the SSF technique has been for the evaluation of beef longissimus tenderness. This has included routine testing of commercial product lines as well as in research. To conduct SSF, immediately after cooking, a 1-cm thick, 5-cm long slice is removed from each steak parallel to the muscle fibers for SSF determinations. The slice is acquired by first cutting across the width of the longest point, approximately 2 cm from the lateral end of the muscle. Using a sample sizer, a cut is made across the longissimus parallel to the first cut at a distance of 5 cm from the first cut. Using a knife that consists of 2 parallel blades spaced 1-cm apart, 2 parallel cuts are simultaneously made through the length of the 5-cm long steak portion at a 45° angle to the long axis of the longissimus and parallel with the muscle fibers.

The 5-cm long, 1-cm thick slice is sheared perpendicular to the muscle fibers using a universal testing machine equipped with a flat, blunt-end blade (Figure 34). The SSF blade is designed to replace the WBSF blade on an automated testing machine. The SSF blade has the same thickness (1.1684 mm) and degree of bevel (half-round) on the shearing edge as WBSF blades and should be used with the same amount of gap (2.0828 mm) for the blade to pass through during shearing. The crosshead speed is set at 500 mm/min to minimize the time required for measurement of shear force. Optionally, SSF could be measured using a WBSF machine equipped with an SSF blade, as described in the “small volume” protocol listed below. In that case, the crosshead speed is dictated by the WBSF machine.

Figure 34.

Slice shear force blade segmenting a meat slice.

Recent data indicate SSF measurement on longissimus and gluteus medius can be conducted with slices obtained perpendicular to the steak/chop cut surface Evidence indicates SSF is much less sensitive to muscle-fiber orientation effects than WBSF.

The limited data available comparing hot vs. cold SSF, however, indicates hot SSF resulted in stronger correlations to WBSF and trained sensory tenderness ratings than cold SSF (Shackelford et al., 1999b). Thus, the implication of more accurate data and the increased convenience of cooking and shearing in the same day leads to a recommendation for hot SSF.

It has been clearly shown that shear force (both WBSF and SSF) does not properly reflect tenderness differences among muscles (Bouton et al., 1978; Harris and Shorthose, 1988; Shackelford et al., 1995; et al., 2004; King et al., 2009). For example, beef biceps femoris and longissimus have similar WBSF values, but longissimus is more tender as assessed by a trained descriptive attribute panel (Shackelford et al., 1995; Rhee et al., 2004). Therefore, it is inappropriate to use shear force to compare tenderness differences among muscles. This likely will lead to misinterpretation of data and false conclusions. Sensory evaluation should be used to compare the tenderness of muscles. When it is necessary to measure shear force in multiple muscles in a given experiment, we urge the investigators not to compare (statistically or otherwise) shear force among muscles and to report the data in such a manner that does not encourage the reader to compare shear force among muscles. Furthermore, we urge the investigators to use a qualifying statement either as a footnote (legend entry) or as a part of the statistical analysis section of the document such as the following sentence from King et al. (2009): “Slice shear force data were analyzed independently for each muscle because measures of shear force do not accurately reflect tenderness differences between muscles (Shackelford et al., 1995; Rhee et al., 2004), because shear force does not accurately represent the contribution of connective tissue to muscle tenderness (Bouton et al., 1978; Harris and Shorthose, 1988).”

Allo-Kramer shear value

A popular alternative to the WBSF method is the Allo-Kramer shear (AKS) method, especially for poultry, fish filets, and further-processed meats (Figure 35). For poultry, the AKS value (kg of force per g of meat) can be determined on surface samples (40 × 20 × 7 mm), 25 mm diameter cores that are cored parallel to the muscle fiber, or strips that are cut parallel to the muscle fiber. Samples should be taken in at least duplicate, and a minimum of 30 breasts should be evaluated per treatment if the data will be published in a peer-reviewed journal. This could be 10 samples per treatment per replication for 3 replications. The instrument should be equipped with a 10-blade AKS compression cell, a 500-kg transducer load cell, a 50, 100, or 200 kg load cell, and a crosshead speed of 500 mm /min (Sams, 1990). Samples must be weighed before shearing. Shear values are calculated by dividing either the total kilogram shear force or total Newton shear force by the weight of the sample (kilograms of shear per gram of sample; or Newtons of shear per gram of sample), and total energy (Joules/g) is calculated as the area under the texturegram. Lyon and Lyon (1991) found that the AKS values had a significant correlation (R² = 0.86) with the sensory texture values of chicken breast meat deboned at various times postmortem.

Figure 35.

Allo-Kramer shear force method: a) Allo-Kramer blades and cell prior to shearing; b) inside of Allo-Kramer shear cell when chicken breast sample (40 × 20 × 7 mm) oriented so that sample is sheared across the muscle fibers.

For further-processed meats, the AKS force is conducted similarly to the description for chicken breast and fish. However, samples will be taken such that they are a specific size. For example, a 25.4 mm diameter core that is 12.7 mm in length is an excellent sample size for this application. A 20 mm cube or sample that is 20 mm in width × 40 mm in length × 7 mm in depth is an acceptable size for use for AKS force of deli meats. The instrument should be equipped with a 10-blade AKS compression cell, a 500-kg transducer load cell, a 50, 100, or 200 kg load cell, and a crosshead speed of 500 mm /min. Shear values are calculated by dividing either the total kilogram shear force or total Newton shear force by the weight of the sample (kilograms of shear per gram of sample; or Newtons of shear per gram of sample), and total energy (Joules/g) is calculated as the area under the texturegram. Both shear force and total energy estimate protein binding strength.

Meullenet-Owens razor shear

Common texture analysis methods, such as AKS, WBSF, and texture-profile analysis (TPA), can be subject to variation due to excising the sample from the whole muscle or variation in texture due to location (Owens and Meullenet, 2010). Hence, there was a need to develop novel methods that can be used on whole samples, especially for poultry, thus reducing labor, time for analysis, and variability due to location. The Meullenet-Owens razor shear (MORS) method was initially developed for poultry meat (Cavitt et al., 2004, 2005a, 2005b; Figure 36). The purpose was to be able to shear intact filets rather than to have to cut samples to an exact dimension. Some advantages of this method are: 1) reduction of experimental error due to sample preparation; 2) the method is independent of sample dimensions; and 3) there is no sample preparation time (Lee et al., 2008). It is important to change the razor blade periodically to prevent dulling of the blade. Recommendations are to change the blade after 100 shear measurements, but it is important to be consistent within a study and make sure that the blade being used is sharp. A blunt version of MORS was developed, which has been used for tougher meat samples (BMORS; Figure 36). Lee et al. (2008) reported that BMORS was more discriminative than MORS at higher shear values, while MORS may be more discriminative for tender meat. With the blunt version, the shear test includes compression as well as shear. This results in slightly higher shear values (energy and force) compared to MORS. The MORS and BMORS are very highly correlated (r = 0.99) with each other using either total energy or maximum force values. Both methods are strongly correlated (r < −0.82; r = −.87 to −0.90 for MORS/BMORS to tenderness intensity, respectively) to consumer sensory attributes such as overall impression, tenderness acceptance, tenderness intensity, and tenderness JAR (Lee et al., 2008). For MORS, a razor blade (8.9 mm wide) penetrates the cooked meat sample to a depth of 20 mm, perpendicular to the muscle fiber. Multiple shears are performed for each filet and averaged. For BMORS, the same blade is used, but the razor edge is cut off, leaving a blunt edge. Replacement of the BMORS blade during testing is not necessary. All other parameters are the same. Cooked meat samples should be used for the assessment of tenderness properties when MORS and BMORS are used.

Figure 36.

The Meullenet-Owens razor shear and Blunt Meullenet-Owens razor shear texture evaluation methods. a) MORS (razor edge) and BMORS (blunt edge) blades with close-up view of edges; b) MORS method with chicken breast filet; c) close-up of shear process as blade shears across muscle fibers. Abbreviations: BMORS, Blunt Meullenet-Owens Razor Shear; MORS, Meullenet-Owens razor shear.

For the MORS and BMORS test, a 20-mm-thick (minimum) piece of meat is placed on a flat aluminum plate with a 1 × 50 mm slot for the blunt blade (0.4 mm wide) to pass through. The texture of the sample is assessed using the BMORS or MORS device on a texture analyzer (TA-YT2i, Stable Micro System, Ltd). While samples less than 20 mm are not recommended, there may be times when larger samples are not possible. In these cases, it is recommended to report the maximum force (N) rather than the total energy. This system has been used with chicken breasts but has also been used for deli meats as described by Luckett et al. (2014). Using the data acquisition software (Texture Exponent 32, Stable MicroSystems, Ltd), 6 parameters are defined: positive force for springiness, positive area for hardness, time to positive peak for cohesiveness of mass, negative force for fibrousness between teeth, negative peak for rubberiness, and gradient to positive peak are obtained. The crosshead speed should be fixed at 5 mm/s, and the sample shear depth at 20 mm with a trigger force of 10 g. A multiblade shear test was developed by Zhang et al. (2021), which showed efficacy when evaluating broiler breast meat with or without myopathies.

Texture-profile analysis methodology: 2-cycle compression

Texture assessment of further-processed meats can be assessed using the texture assessment methods defined above. However, assessment of physical properties in further-processed meats is needed to understand texture. A common method of texture assessment is the TPA method. This method entails 2 sequential compressions on the same sample (Bourne, 1978; Montejano et al., 1985; Claus, 1989) and is classified as a dual-compression test. Measurements obtained using TPA that are commonly applicable to processed meat type samples include: hardness, cohesiveness, springiness, and chewiness (Table 24). Less commonly measured is adhesiveness. Hardness is measured with a single compression test, and the other 4 measurements are from the dual-compression test. Proper setup for performing TPA entails some pretesting of the samples that will be analyzed so that the instrument is correctly configured. Using a capacity load cell that is too high a may not be sensitive enough for delicate samples. With a too-low-capacity load cell, some samples may exceed its capacity. The following are key elements/steps that should be observed.

A standard configuration sample should be predetermined. In general, the shape of the sample should be such that it retains its basic orientation during compression. For core samples removed from a sausage, the diameter of the core should be at least equal to the height of the sample, but preferably a little larger. The core should be removed perpendicular to the cut surface. For sausages that are not to be cored, the sections should be cut out perpendicular to the long axis, again with the samples being wider than they are tall. If the intent is to perform TPA on non-cored samples so that the contribution of the skin (coagulated protein) can be included, this can be done if the sausages have a fairly equal diameter among the products being evaluated. If not, then cores should be used, and an alternative test for the skin contribution (e.g., puncture test) should be used. Samples should be presented to the TPA by being placed on their flat cut surface. Important aspects of evaluation include: 1) the temperature of the samples at the time of compression needs to be standardized and documented; 2) the compression platen (plate) that is selected must be large enough so that when the sample is compressed to the selected final compression percentage, none of the sample is outside the outer edge of the plate; 3) ideally select a load cell in which the maximum peak force of the sample will fall in the middle of its capacity; 4) the texture analyzer should be calibrated before testing; and 5) set the height of the compression platen (plate) adequately above the sample so there is room to place and remove the sample as well as inspect the surface of the compression plate for residual sample post compression. TPA instruments vary in parameter settings and therefore need to be specified. Settings that should be documented include: 1) pre-test, test and post-test speed, usually set at 5 mm/s; 2) trigger force is the threshold for the instrument to recognize contact with the sample; 3) fracture detection parameters; 4) delay time between compressions; 5) percentage compression usually set at 75%, 70% or 60% of sample height.

Careful attention should be given to how the instrument designates measurement criteria. The word descriptions of the TPA traits help provide a general understanding of the physical measurements. However, in the literature, there are a few variations on how some of the parameters are actually calculated. The more traditional method of calculating the major traits is detailed first (Table 24).

Figure 37 shows a typical 2-cycle TPA compression curve in Newtons. Hardness, which is the peak force (N) of the first compression, relates quite well to sensory firmness. It is important that this first compression does not break the product, since it will prevent the accurate measurement of other items. Cohesiveness is determined by independently calculating the integrated area under each compression curve. The area under the first compression curve (A1) starts from the initial contact (point P1) with the sample up to the completion of the designated percentage compression (P2). Note that it was not defined as the point at which the maximum peak force was achieved. The reason for this will be made apparent when discussing a sample that has a significant fracture point. The integrated area under the second compression (A2) starts when the compression plate comes in contact with the sample (P3) and concludes when the same downstroke distance of the first compression has been achieved (P4). Cohesiveness is then represented as a ratio A2/A1 (no units). In some cases, cohesiveness is presented as a percentage (A2/A1 * 100).

Figure 37.

Example of a texture-profile analysis 2-cycle compression curve. Abbreviation: TPA, texture-profile analysis.

Springiness is determined as the base width of the second compression (b2 = P4−P3, mm). Since comparisons are often made between products, using only b2 for springiness is sound if the exact sample height is controlled. To account for some variation in sample height, springiness can be calculated as b2/b1, which then puts it on a ratio basis rather than an absolute recovery height measured basis. In some cases, springiness is presented as a percentage (b2/b1 * 100).

Chewiness is the product of hardness times cohesiveness times springiness (N × mm). In many respects, the percentage compression used to determine hardness should be greater than that used to determine cohesiveness and springiness. This means a sample from the same product should be compressed more extensively (e.g., 75% of the sample height), and then a second sample should not be compressed as much (e.g., 60%). For hardness, the sample should be more fully compressed so that it is indicative of what a sensory panelist evaluates for firmness. If the sample is over-compressed, there may be so much structural damage from the first compression that the second compression may only show a minor compression curve. This is why pretesting samples is of value to determine an applicable compression percentage. Nevertheless, if a direct comparison needs to be made to a published paper, the same percentage compressions should be utilized.

Another example of a 2-cycle compression curve is illustrated in Figure 38. In this example, the sample peak during the first compression did not correspond to the end of the downstroke (specified percentage compression of the sample height). The end of the compression downstroke occurred at P2. Users are cautioned to be sure that they know where the completion of the downstroke occurs for integrating the area under the curve.

Figure 38.

Two-cycle compression test with 2 peaks during the first compression. Abbreviation: TPA, texture-profile analysis.

An alternative calculation for TPA cohesiveness that is sometimes used is to integrate the total energy under the curve defined as A1 and A2 (Figure 39). In this method, cohesiveness is calculated as A2/A1. Note that the endpoints of P2 and P4 represent points where the curve goes to zero or levels off after the compression cycle has completed. This approach avoids the need to validate the completed downstroke position as discussed in Figures 37 and 38.

Figure 39.

Example for calculating alternate cohesiveness using texture-profile analysis. Abbreviation: TPA, texture-profile analysis.

TPA users need to fully disclose how they calculated all traits, and in particular, values for chewiness will be difficult to interpret without knowing if cohesiveness was calculated as a ratio or percentage, or if springiness was only based on the width of the second compression, a ratio (b2/b1), or a percentage (b2/b1 * 100). For ground beef patties, TPA can be conducted using one core (2.54-cm diameter) removed from the center of 10 cooked patties and compressed twice to 70% its original height to determine hardness (N), cohesiveness (ratio), springiness (height sample recovers, mm), gumminess (product of hardness times cohesiveness, N), and chewiness (N x mm). A crosshead speed of 100 mm/min should be used. AKS force can also be used. Using AKS, 1 strip (2.5-cm wide) should be cut from the center of each of 10 patties per treatment and replication. Each strip should be analyzed by either shearing one with a multiblade Allo-Kramer shearing device or shearing 3 times with a single-blade, straight-edge Allo-Kramer shearing device. Texture has also been assessed where these strips are sheared 3 times with a straight-edge SSF blade attachment. In each situation, a crosshead speed of 200 to 250 mm/min should be used for an automated testing machine.

Texture-profile analysis methodology: 1-cycle compression

One-cycle compression tests can be used to understand the underlying structural component of meat products. While it is used similarly to 2-cycle compression tests, its use is less common, and there is limited information compared to the 2-cycle TPA test. The test should be conducted as defined for the 2-cycle TPA test; however, only one compression is used. Data collected are peak force, A1, B1, P1, and P2 as defined in Figure 37. Not all attributes can be calculated in the one-cycle test as defined for the 2-cycle test.

Protein binding strength can be determined using a procedure that is modified from Field et al. (1984), as illustrated in Figure 40a. For this application, the diameter of the deli meat or ground meat patty should be between 10 and 12.5 cm in diameter. However, a machine shop could make an attachment specific to fit texture analysis instrumentation that could be used for meat pieces with other dimensions. This device was designed for 12.7-mm-thick deli slices, but could be used for products with other thicknesses. The thickness of the slice should be between 3.2- and 12.7-mm thick and should be the same for all samples. A 25.0-mm diameter steel ball (chrome alloy grade 25) is attached to a rod that is attached to the texture assessment instrument. Nails are placed manually through each sample into 1.6-mm holes drilled on the top of a plexiglass stand to secure the deli meat slices during testing. A minimum of 4 slices should be evaluated from each treatment for each replication due to biological variability in meat products. Nail holes were drilled 0.5 mm apart in 1 mm deep holes around a circle with a nail diameter of 4.5 mm and an inside diameter of 4.0 mm. The plexiglass stand is placed on the flat, circular surface of the texture analysis instrument. The slice should be aligned so that the steel ball penetrates the middle of the meat slice. The steel ball is positioned directly above the meat slice, and protein bind is estimated as the peak force (N) for the polished steel ball to burst through the slice of deli meat at a speed of 100 mm/min. If the meat slice breaks in a place other than the center of the slice, that data should not be used (Daigle et al., 2005).

Figure 40.

Ball head (a) and star probe (b) used in compression tests.

A modification of the penetrating probe/steel ball test was developed to determine if there was enough functional protein in a processed meat product. For example, this test can be used to determine if mechanically separated poultry meat has sufficient protein binding in Newtons (N) to form a frankfurter. This testing was developed to evaluate if mechanically separated chicken (MSC) that was treated with antimicrobials to control Salmonella was not denatured by the antimicrobial dips and sprays that were used. To conduct this test, MSC is weighed and preblended with 1% NaCl. For each treatment, six 11.5 cm × 11.5 cm patties that weighed approximately 180 ± 0.1g should be formed and baked to an internal temperature of 74°C on a broiler pan with a rack (Viking Professional, Greenwood, MS, USA) at 204°C. Similar instrumentation could be used for other-sized patties, but these sizes work well with the equipment that is available. After reaching a temperature of 74°C, samples are removed from the oven and cooled to room temperature (22 ± 2°C). The instrument that is used to measure texture should be equipped with a 500 N load cell and a 25.0-mm diameter steel ball (chrome alloy grade 25). For each sample analysis, a patty is centered on the plexiglass stand so that the steel ball penetrates the middle of the sample once. Compressive load at maximum peak force (N) estimates protein binding value. Values greater than 20 N generally have sufficient protein to bind to form a frankfurter with acceptable water-holding capacity and texture.

Star probe

The star probe was developed as an alternative to WBSF and TPA evaluation for pork and further-processed meats (Figure 40b) and has a 5-pointed cutting edge. The sharpness of the blade edges should be maintained. Crosshead speed is up to the user and should be defined. For all products, the sample needs to be sufficiently thick so that the star probe can fully penetrate the surface without contacting the base that the sample rests on. Multiple probe measurements can be made on the same sample, provided that there is sufficient space between each penetration. For intact muscle cuts, a minimum of 2.5 cm between penetrations should be used. The star probe can be used for finely comminuted sausages. Investigators need to determine the minimum distance between penetrations that does not alter the structural integrity of the subsequent penetration. The star probe should not be used on ground patties as they tend to be more fragile, and the data will not provide a true measure of penetration.

Tensile testing

Tensile testing is a method that provides information on the strength of the bind within a further-processed meat product (Honikel, 1998). To conduct this method, one tensile grip should be attached to the base of the instrument that is used to analyze texture, while the other grip should be attached to the load cell. Initial grip separation can be variable, but 28.0 mm has been recommended by Honikel (1998). Crosshead speed should be approximately 1.0 mm/s (50–60 mm/min until rupture). This is based on a strain rate of approximately 2 per min. Therefore, if the strip is 28 mm in length, an extension rate of 56 mm/min is recommended. Honikel (1998) recommended a 28-mm-long strip that is cut out in the shape of a dumbbell, such that there is a 4 to 1 to 0.5 length to width to thickness ratio for the sample. For example, at 28 mm length, there should be a 7 mm diameter in the interior portion of the dumbbell (18 mm in length), and the entire strip should be 3.5 mm thick. The outer ends (5 mm long on each side) of the dumbbell should be approximately twice the width of the interior portion of the dumbbell. In addition, the sample should be continuously cut so that there is a smooth sample parallel fiber direction. The purpose of the dumbbell shape is such that the sample tears in the center of the strip and not on the ends. If the meat sample is not susceptible to tearing in the middle, the dumbbell shape is not necessary. For the test to be acceptable, the sample needs to break in the thinner portion of the dumbbell, in the center, such that it occurs in the parallel-sided region of the sample. A minimum of 8 samples should be evaluated per treatment, and only samples meeting the above criteria should be accepted as data. Breaking stress should be reported as N/m² (Pa).

The following calculations are often reported: Breaking stress (N/m² (Pa)) = (peak force)/(measured width x thickness) s; Breaking strain (extension of peak force/sample length) is the relative elongation of a material at its breaking point and is unitless.

Total energy to fracture (Joules) = area under the curve.

Blunt rod for skin assessment

The skin strength of further-processed meats can be affected by processing, cooking, ingredients, and other factors. Samples are prepared as defined for penetrating ball testing, except that a blunt rod is used to penetrate the skin of the sample. Samples can be evaluated cold or hot as long as a consistent temperature is maintained. Penetration is defined as when the peak force is reached, and the curve shows failure. Data are reported as peak force in N or kg. Blunt-end rods should be of sufficient circumference to provide reasonable measurement of the penetrating force, but not so large that the force is more indicative of compression. The diameter of the blunt-end rod is dependent upon how delicate the skin is on the processed meat and the capacity of the load cell installed on the texture analyzer.

Slice integrity

Slice Integrity is used for sliced deli meats to understand protein binding within the slice. For this method, each deli product treatment is sliced into 100 1.6 mm slices. Slice integrity is determined by the number of slices out of 100 that remain intact (no holes or tears) after slicing and holding in the air by fingertips to ensure that it does not tear before packaging. Slice integrity is an indicator of protein binding.

Torsion testing

Torsion testing is a useful texture analysis method for use on finely comminuted sausages and studies on the fundamental properties of protein gel systems. As the name implies, the sample is mounted in a device that applies torsional force as a result of twisting the sample until the point of failure (Figure 41a). This method yields 2 independent mechanical properties, which are stress (kPa) and strain. Stress is the peak force that occurs at the point of failure. Stress is sometimes also referred to as rigidity or fracture stress. Strain is unitless but can be thought of as the distance or time necessary to reach the failure point. Strain has also been described as deformation, that is, the extent to which the sample has to be deformed to reach its failure point. Cooked sausages or protein gels that are tested using a torsion gelometer must be homogeneous and isotropic. Isotropic is a key feature in that no matter how the sample was removed or obtained from a cooked sausage, the sample mounted on the torsion device must be independent of orientation. As such, finely comminuted processed meats adhere to this requirement. Frankfurters and bologna-type sausages are compatible with this method. In the case of frankfurters, a standard section from the sausage is cut perpendicular to the length (Figure 41b), and then Mylar discs with alignment notches are glued (cyanoacrylate) on each end. The sample is then milled (Figure 41c, Accu-Tool, LLC, Apex, NC, USA) to an hourglass shape to a standard diameter (10 mm). This, of course, removes the coagulated skin layer from the sausage and assures that the failure will occur independent of the device that holds one end stationary while the other end is twisted. Basically, any finely comminuted sausage or protein gel that the milled hourglass-shaped sample can support itself (stand erect; Figure 41d) will enable torsion testing.

Figure 41.

Torsion testing with comminuted meat products (a) where the whole system is set up, (b) cutting and gluing a frankfurter section, (c) a core being milled using a Accu-Tool, (d) to an hourglass shape (d) to a standard diameter (10 mm).

Spurious results may be obtained with this texture method if there are significant air pockets during the sausage manufacturing process. If an air pocket is present in the narrow portion of the dumbbell-shaped sample, it will prematurely fail (tear). Obviously, samples with visual air pockets on the outside of the milled sample should be excluded from use. However, after the sample has been processed on the torsion device, the interior of the sheared surface should be assessed for flaws.

Instrument calibration

Calibration is essential with any instrument, and verification is recommended at 12- to 18-mo intervals. Calibration is the daily spot-check of the instrument’s accuracy. It is performed by placing a known weight on the transducer or by applying a known voltage to the cell via a shunt. Adjustments can be made so that the instrument output matches the known weight or voltage input. The calibration procedure is usually performed using a single value or with weights at approximately 20% and 80% of the force values expected in the test (see ASTM Standard E4-24, 2010, for other approaches). Some manufacturers recommend that the shear blade attachment be in place during calibration so that its weight and any side drag are accounted for during taring. Many protocols calibrate and then tare the load cell. The calibration procedure should be defined and followed during testing. Automated testing machines should be verified according to the manufacturer’s instructions. When crosshead speed is critical to the results, it is recommended to have it verified as well.

When force values form the basis of value, for example, premiums for tenderness, instruments will need to be validated daily (ASTM Standard F2343-06, 2006). This process is referred to as validation, a procedure that documents, confirms, and provides assurances that force measurements will consistently meet predetermined specifications and force attributes. Further, validation also provides assurances that instruments are installed and operated within the manufacturer’s specifications (ASTM Standard F2341-05, 2005).

The USDA’s Livestock, Poultry and Seed Program approves third-party laboratories for conducting SSF/WBSF testing to ensure that the requirements of an ASTM International tenderness standard have been met. Third-party laboratories are approved to conduct SSF and/or WBSF through proficiency testing conducted in conjunction with an AMS-designated SSF/WBSF reference laboratory (USDA Livestock and Poultry Program, 2012).

Data preparation

Sensory data are traditionally either collected using a handwritten ballot filled out by each panelist or collected into a computer software program. Regardless of the collection procedure for the data, it needs to be organized so that one row represents the responses of one panelist for one sample. Variables for sensory day, session within sensory day, order, 3-digit sensory codes, panelists, treatment codes, and sensory variables for each response should be included as columns. Variables that might be potential covariates or blocks, such as cook time and cook yield, should be included.

If data have been entered into a spreadsheet with panelists’ results for each sample reported, then reorganization is not necessary, and the next step of analysis can begin. If data have been collected using a computer program, the data might need to be unscrambled from the randomized sample sheet to assign the observed values to the actual treatments using a program like Unscrambler from SAS or within Excel (Microsoft). If handwritten ballots are used, data should be entered into a spreadsheet and then verified by independent personnel using the original data sheets. Computerized data sheets should be checked for layout and format. If a panelist is not present for a sensory day or sample, the missing data should be indicated by a “.” or the cell left blank, depending on the statistical analysis program being used. If a panelist evaluates an attribute and does not find it present, a numerical value of “0” should be entered into the cell if 0 = none or absence of the attribute. Summary statistics for mean, standard deviation, standard error, minimum, and maximum values should be calculated either within the spreadsheet or by a data analysis program. Summary statistics can be useful in conducting diagnostics to ensure that data are within range and that gross errors in data entry are found.

It is assumed that researchers have experience using the statistical package for data analysis of their choice. Many statistical packages are available for use, such as SAS, SPSS, R, JMP, XLSTAT, and others. Resources to understand programming language should be used to implement the data analyses, and will not be covered here. If a data program, like Compusense, is used, it is the responsibility of the sensory professional to understand how the data is being analyzed and to verify that the model and experimental design match.

After assurance that the data are correctly entered, tests for normality are conducted. Consumer data are commonly abnormal in distribution, but all data should be tested for normality because normality is an assumption when conducting ANOVA. Data can be analyzed with non-normal assumption procedures like PROC GLIMMIX of SAS that generalize the MIXED procedure. The GLIMMIX procedure generalizes the MIXED and GENMOD procedures in 2 important ways. First, the response can have a non-normal distribution. The MIXED procedure assumes that the response is normally (Gaussian) distributed. Second, the GLIMMIX procedure incorporates random effects in the model and thus allows for subject-specific (conditional) and population-averaged (marginal) inference. If data are not normally distributed and the data are analyzed using a model that does not account for non-normal distributions, the data should be tested before analysis. Normality of data can be tested in many ways. Some statisticians recommend plotting frequency distributions of the data and examining the shape of the curve. If the curve is bell-shaped, the data are normally distributed. This method, however, is not highly accurate. Most statistical programs have methods to test for normality. In SAS, the Proc Univariate function performs many descriptive statistics for normality, such as Q-Q plot, stem-and-leaf plot, box plot, and normal-probability plot. This procedure also conducts Kolmogorov-Smirnov, Shapiro-Wilk, Anderson-Darling, and Cramer-von Mises tests to evaluate normality. The Box-Cox transformations in SAS can be used to determine logarithmic transformation constants that can be used to convert the data to a normal distribution. Other transformations also might be appropriate. If data are transformed—for ease of interpretation—least-squares or unadjusted means should be transformed back to the original scale, but least significant difference values from the original scale cannot be used. The researcher should also check to ensure that the back-transformed means are not substantially lower than the means in the original data. This can be an issue when the transformation is logarithmic.

A second way to test for normality is to plot the residuals of the data. It is suggested, however, that if there are missing observations, a “means” statement may not be used, and the researcher should use least-squares means to compare treatment means. In sensory data, it is not uncommon to have a panelist missing from a day of evaluation; in that case, least-squares means should be used. Diagnostics are run to ensure that data comply with the assumptions of an ANOVA; i.e., model errors are normally and independently distributed random variables with mean = 0 and variance = σ², constant variance and homogeneously dispersed, and the measurements were acquired in a random order in a completely randomized design (Huntsberger and Billingsley, 1979; Steel and Torrie, 1980; Devore and Peck, 2005). To run diagnostics, the predicted and studentized residual values of the data need to be calculated and obtained from the ANOVA procedure. The predicted and studentized residual values can be plotted as in Figure 42. For example, the “gplot” procedure in SAS will plot the studentized residuals by the predicted observation values to determine whether or not the variance is constant and homogeneously dispersed. If the variance is constant and homogeneously dispersed, then the plot should look similar to the scatter plot in Figure 42, where there is no pattern to the residuals. Normality of the data can be assessed using the “gchart” procedure, and if the data are normally distributed, then the histogram should somewhat resemble a bell-shaped curve. It should be noted that the ANOVA is not very sensitive to departures from normality or unequal variance and is a relatively robust statistical test (Ott and Longnecker, 2001). Slight deviations from normality will most likely not affect analyses and interpretation of results. If there is concern, a statistician should be consulted.

Figure 42.

Example plot of residuals (a) to evaluate whether variance is constant and homogeneously dispersed and (b) histogram to evaluate normality of the data.

Significance levels for analysis should be predetermined. While an α-level of < 0.05 is commonly used, in a small dataset, an α of < 0.10 could be justified, whereas in a large dataset, an α of < 0.01 might be more appropriate. It is important to predetermine significance levels based on experimental design and observation values before initiation of the study to prevent potential biases when interpreting sensory data. In addition, α-error should always reflect the level of confidence that the researchers will require to make inferences before any data are collected or analyzed.

Descriptive sensory data analysis

Most descriptive sensory data that is generated using trained panels are analyzed using ANOVA, where each sensory attribute defined in the ballot is evaluated for the effects of treatment independently. This type of analysis provides the opportunity to understand if specific attributes differ among treatment levels. For trained descriptive data, an initial model that includes random (sensory day, session, and order) and fixed (treatment and subsequent interactions) effects should be developed to reduce the risk of encountering β-error. Potential covariates should be included in the model if they help to explain the error. Random variables that account for additional variation, and to make inferences about the population from which the experimental units were derived, can be included. In meat science data, slaughter day, processing day, shift of production, and plant where experimental units were collected are examples of random variables. Meat scientists many times conduct repeated-measure experiments, especially when conducting studies that include different storage times or when multiple steaks are removed from a subprimal and then randomly assigned to treatment. It is wise to consult with a qualified statistician to ensure that the analysis is being conducted correctly and that the error terms that are used to test effects are appropriate.

Consumer data analyses

For consumer data, “Consumer” is included in the model since consumers normally account for the greatest source of variation. Consumers are untrained, and variation associated with consumers indicates differences in consumer perception or liking. Consumers can be evaluated using multivariate tools to understand potential segmentation or categorization of consumers that may help in explaining their source of variation (see discussion below).

When analyzing consumer data, it is also important to look at the percentage of “Like Extremely” and “Like Very Much” ratings (top 2 boxes in a 9-point hedonic scale) as well as “Dislike Extremely” and “Dislike Very Much” ratings (bottom 2 boxes) to check for shifts in rating patterns among treatments. It is possible for mean scores to not be significantly different but have shifts in top 2/bottom 2 box ratings, indicating consumer segmentation or polarized ratings.

Consumer data should be analyzed similarly to descriptive sensory data, where covariates, fixed and random effects, model considerations (repeated measures), and interaction effects are included in a full model. Consumer is included as an effect in the model and, in most cases, contains the highest source of variation. Location, if there are multiple locations, is included as fixed effects and is used to understand if there are differences in consumer sensory perceptions across locations. An example would be if consumers from the west, east, and southeast evaluated beef steaks. This partition in the model would assist the sensory professional in understanding if there were regional differences in consumer perception. Random effects of sensory day, session (if more than one session is conducted per day), and order are included in the full model. These effects account for variation in the model but are usually not used to interpret findings. The full model can be defined, and interaction effects may or may not be included in the model.

Data are presented as discussed for descriptive sensory data, but using consumer attributes. It should be noted that in all sensory data, least-squares means or means should be reported to 1 decimal place more sensitive than the original data (i.e., if the panelists are evaluating an attribute to the nearest 1 unit (1, 2, 3, …), data should be presented to the nearest 0.1 unit) and the measure of variance should be reported to 2 units more sensitive.

Multivariate analyses

Sensory data are by nature multivariate. When a human evaluates a sample, activated sensory receptors send information to the brain. Trained sensory panelists are proficient in either ignoring information (i.e., ignoring information that would be considered based on visual appearance and its relationship to flavor), segmenting or pulling out a specific attribute from the sensory stimulus (i.e., juiciness is evaluated independently of muscle-fiber tenderness or beef-flavor identity is measured independently of cooked beef fat), and quantifying it. Consumers might rate their overall liking of a sample and then be asked to evaluate specific attributes, such as their overall liking or intensity of juiciness and/or tenderness. Univariate analysis of independent attributes provides for the interpretation of treatment effects. Key attributes that differ by treatment provide researchers the ability to determine if treatments differ or not. Whereas multivariate analyses provide the opportunity for understanding how multiple variables are impacted by treatments and how consumer and trained sensory data might be related. These can be very powerful tools in understanding whether treatments affect the sensory properties of meat samples. Multivariate analysis tools provide additional information. If the dataset does not have relevant information, such as differences in ANOVA and variation is very low, however, multivariate techniques are not a way to gain information.

In using multivariate techniques, the number of observations is not as important as the strength of the relationship. For example, if a quantitative attribute used in trained panel analysis, such as cooked beef-flavor identity, does not have a quantitative relationship in univariate analysis, then these data will not be useful in multivariate analyses. There are useful resources that discuss multivariate analysis—Meilgaard et al. (2025) and Lawless and Heymann (2010). Many of the statistical packages have training that can also assist in understanding how to conduct these analyses. It is assumed that individuals using this guideline are familiar with these or comparable resources. A general description of common multivariate tools that can be used in sensory data, how they are used, what type of information they provide, and a general interpretation of data are discussed below. A dataset was extracted from Miller et al. (2023). These data include beef samples (16 treatments) that were selected to differ in flavor due to quality grade, cut, cooking method, and cooked internal cook temperature endpoint. These samples were evaluated by a trained descriptive flavor attribute panel using the Adhikari et al. (2011) method, by consumer sensory evaluation in 4 different cities (n = 80 per city), and for volatile flavor chemical analyses. Not all of the data or attributes were included in these analyses, but the data were used to illustrate how to use multivariate techniques to understand relationships within these data.

Principal component analysis

Principal component analysis (PCA) helps understand relationships among variables. Like correlation analysis, PCA studies the orthogonal relationships among variables in a dataset, but it also allows the visualization of how variables move in relation to one another. In PCA, orthogonal variation in the dataset is explained by combining variables that are related into a set of new variables, called principal components, which are uncorrelated and orthogonal to one another. The first principal component (PC 1) is derived to account for the greatest amount of variation; subsequent factors account for progressively less variation. The number of factors that can be derived is equal to n − 1, where n equals the number of variables in the dataset. Generally, only the first few factors explain significant portions of the variation, and, thus, only the first few are scrutinized. The percent of the variation that is accounted for by each factor is provided in the analysis. A bi-plot is generated that shows the relationships between factors, usually factors 1 and 2, and the sensory variables and treatments.

Figure 43 is a bi-plot generated from the aforementioned data (Miller et al., 2023). The objective is to understand relationships between beef-flavor descriptive attributes and the 16 treatments. The original data (n = 640) were averaged across panelists and treatment so that there were 16 lines of data used for PCA analysis. The first factor, or Principal Component 1, accounted for 47.3% of the variation, and the second factor, or Principal Component 2, accounted for 22.97% of the variation. The third component and subsequent components were not significant and were not presented. To interpret the bi-plot, attributes and treatments are plotted. If attributes are in the center of the plot, they are not contributing to variation in the 2 principal components. Attributes or treatments that are in a similar area are highly related. Attributes or treatments near the 0 on the vertical axis, but located away, either negatively or positively, from the horizontal axis, are accounting for more of the variation in Principal Component 2. Conversely, attributes or treatments near the 0 on the horizontal axis, but located away, either negatively or positively, from the vertical axis, account for more of the variation in Principal Component 1. Variables in the 4 quadrants between axes account for some of the variation, either negatively or positively, for both principal components. Variables used in the PCA should be measured using the same scale; these data were collected using a 16-point Spectrum^™ Universal scale. To interpret the bi-plot, treatments clustered with variables are related. Therefore, Choice and Select bottom round roasts cooked in a crockpot to 80°C and top sirloin steaks grilled to 80°C had similar umami and beef identity. Liver-like flavor was related to Choice top loin steaks cooked on a George Foreman grill to an internal cook temperature of 80°C, whereas metallic flavor was most closely related to top sirloin steaks cooked to 58.3°C on either a George Foreman or grill. Other relationships are apparent but will not be discussed. Data where treatment may be represented by different products across companies could be analyzed for either trained or consumer sensory attributes to understand how these products differ in either trained descriptive sensory attributes or consumer attributes.

Figure 43.

Principal component bi-plot of trained, beef descriptive flavor attributes from the beef-flavor lexicon (red dot) and 16 treatments (blue square), where TL, top loin steaks; GF, George Foreman grill; CTL, Choice top loin steaks; SBR, select bottom round roasts; CP, crockpot cooked; CBR, Choice bottom round roasts; and TS, top sirloin steaks. Adapted from Miller et al. (2023).

Partial least-squares regression

Partial least-squares (PLS) regression is a one-step iterative process that computes latent vectors to explain both independent and dependent variables. This tool can be used when more than one dependent variable is being predicted or when independent variables are measured using different scales. For example, if trained and consumer sensory responses were obtained on a set of samples, the trained sensory responses could be defined as the independent variables and the consumer responses defined as the dependent variables (variable being predicted so that a trained panel could be used to measure a product and understand how consumer responses would be affected). A bi-plot is shown in Figure 44, which was calculated from consumer and trained descriptive attribute sensory data for the same data presented for PCA analysis. Note that this bi-plot presents the relationships between consumer responses and trained descriptive sensory flavor attributes adapted from Miller et al. (2023).

Figure 44.

Partial least-squares regression bi-plot for trained descriptive sensory attributes (red dot) to predict consumer sensory attributes (blue square), adapted from Miller et al. (2023).

These data assist the researcher in understanding what trained panel attributes to target for evaluation or that are most likely related to overall consumer liking. Once a relationship between consumer responses and trained panel responses is established, trained panel evaluation can be used to evaluate additional products at a lower cost. In this analysis, the experimental unit is the sample evaluated by each consumer and the trained descriptive panelist; a higher n is used to understand these relationships (n = 451 for the same data presented in Figure 44). The variation defined by the model is defined as 20.2% of X, which explains 45.5% of the variability in Y for the first 2 dimensions. The type of information obtained from PLS bi-plots examines variables that cluster. For example, from the bi-plot presented in Figure 44, it is apparent that consumer attributes of grilled flavor, flavor, and beef-flavor liking were highly related to overall liking. Additionally, liver-like and bitter were not related to overall consumer liking. Brown/roasted tended to be most closely related to overall consumer liking and sour basic taste, while closely related to the consumer liking attributes were most closely related to grilled flavor liking. The grouping of attributes provides an understanding of interrelationships within trained and consumer attributes and between these attributes. Different data can be used in PLS analyses, such as understanding chemical measures and either consumer or trained panel descriptive attributes.

Preference mapping

This statistical tool links consumer acceptance to trained sensory attributes and has become extensively used in the industry. The data are converted so that they are orthogonal. It is a powerful statistical tool to determine how trained panel results are related to consumers’ overall liking ratings. Preference mapping provides a rapid, objective, and reproducible tool for making product decisions. There are 2 main preference mapping tools: internal preference mapping and external preference mapping. Each tool will be discussed to provide information on how to use these tools and how to interpret these analyses.

Internal preference mapping (IMP) uses hedonic data and uses a variation of PCA. The data is converted so that they are orthogonal. On the bi-plot, instead of variables from trained panelists, consumer data and treatments are plotted. This method assists the researcher in understanding the relationship between consumer clusters or groups and overall liking and/or treatments. If consumers are clustered with a treatment, they have a preference for that treatment. An example of an IMP bi-plot is presented in Figure 45. Note that Principal Component 1 accounted for 88.79% of the variation, and Principal Component 2 accounted for 6.70% of the variation. These data are for the same dataset used for Figures 43 and 44, except that the data are presented where consumer response is the experimental unit. Consumer clusters show that not all consumers have the same preferences. In Figure 45, there are 4 general consumer clusters—one in each quadrant. Consumers in the upper right quadrant have a stronger preference for grilled flavor liking, whereas consumers in the left quadrant were furthest away from overall liking. The sensory professional could use these data to further understand why consumers are segmented as defined by pulling these consumers from the dataset and further examining their responses and/or other information obtained from these consumers, such as demographic data, one-on-one interview data, etc.

Figure 45.

Internal preference mapping bi-plot for consumer overall liking attributes (red dot) and individual consumers (blue diamond). Adapted from Miller et al. (2023).

External preference mapping

External preference mapping (EPM) is used for product optimization and in product development. This tool correlates consumer preference data and trained sensory attributes and/or chemical data. Information from EPM assists the researcher in understanding what trained sensory attributes or chemical characteristics drive or are highly related to consumer likes and dislikes. The bi-plot is generated using PLS as the data were measured using different scales. Also note that the data-defined consumer is an experimental unit. While the same dataset was used in Figures 43 and 44, fewer data points were used for simplicity. Attributes are mapped in relation to consumers’ overall like ratings (Figure 46). These data show that bloody/serum-like was most closely related to consumer overall liking. However, pentanal, hexanal, bitter, and umami were negatively related to consumer overall liking. Consumers from 2 different quadrants could be identified, and further information on what was positively and negatively driving consumer acceptability, could be. These data provide direction for products that are higher in the positive and lower in the negative flavor attributes and help in understanding volatile chemical compounds that may be driving flavor.

Figure 46.

External preference mapping bi-plot for consumer overall liking (green dot), trained descriptive flavor attributes and volatile flavor chemicals (red dot), and consumers (blue diamond). Adapted from Miller et al. (2023).