Research was conducted at Charles Sturt University-Riverina, Australia
Optimizing the sensory characteristics and acceptance of canned cat food: use of a human taste panel
Article first published online: 27 FEB 2008
© 2008 The Author. Journal compilation © 2008 Blackwell Verlag GmbH
Journal of Animal Physiology and Animal Nutrition
Volume 93, Issue 1, pages 52–60, February 2009
How to Cite
Pickering, G. J. (2009), Optimizing the sensory characteristics and acceptance of canned cat food: use of a human taste panel. Journal of Animal Physiology and Animal Nutrition, 93: 52–60. doi: 10.1111/j.1439-0396.2007.00778.x
- Issue published online: 7 JAN 2009
- Article first published online: 27 FEB 2008
- Received: 21 March 2007; accepted: 21 September 2007
- Felis catus;
- feeding behaviour;
- human sensory evaluation
A methodology based on descriptive analysis techniques used in the evaluation of human food has been successfully refined to allow for a human taste panel to profile the flavour and texture of a range of cat food products (CFP) and their component parts. Included in this method is the development of evaluation protocols for homogeneous products and for binary samples containing both meat chunk (MC) and gravy/gel (GG) constituents. Using these techniques, 18 flavour attributes (sweet, sour/acid, tuna, herbal, spicy, soy, salty, cereal, caramel, chicken, methionine, vegetable, offaly, meaty, burnt flavour, prawn, rancid and bitter) and four texture dimensions (hardness, chewiness, grittiness and viscosity) were generated to describe the sensations elicited by 13 commercial pet food samples. These samples differed in intensity for 16 of the 18 flavour attributes, which allows for individual CFP flavour profiles to be developed. Principal components analysis (PCA) could successfully discriminate between samples within the PCA space and also reveal some groupings amongst them. While many flavour attributes were weakly correlated, a large number (describing both taste and retro-nasal aroma qualities) were required to adequately differentiate between samples, suggesting considerable complexity in the products assessed. For both MC and GG, differences between samples for each of the texture dimensions were also found. For MC, grittiness appears to be the most discriminating textural attribute, while for GG viscosity discriminates well between samples. Meat chunks and gravy/gels differed significantly from each other in both flavour and texture. Cat food products differed in their liking ratings, although no differences were found between homogeneous, MC and GG samples, and eight flavour attributes were correlated with overall liking scores. It is now necessary to determine the usefulness and limits of sensory data gathered from human panels in describing and predicting food acceptance and preference behaviours in cats. For instance, while the sense of taste in cats appears generally similar to that of other mammals, they lack a sweet taste receptor (Li et al., 2006), which may limit the applicability of sweetness ratings obtained from humans. Modification of existing techniques used with human food research, such as external preference mapping (Naes and Risvik, 1996) may be useful. Ultimately, this may facilitate more economical and efficient methods for optimizing cat food flavour and texture and predicting the effects of composition and processing changes on cat feeding behaviour. This will require collaboration between pet food manufacturers and nutritionists, animal behaviourists and human sensory scientists. The results of this preliminary study should assist in this process.
Pet food production is a highly competitive, multi-million dollar industry, and represents a significant share of the internationally prepared food industry. In urban environments, the essentially carnivorous domestic cat (Felis catus) and dog (Canis familiaris) can exercise relatively little choice as to what and how they are fed. There is therefore a considerable responsibility on pet food manufacturers to develop products that are both nutritious and palatable. In addition, they should be convenient to use, economical, and acceptable to the owner (Booth, 1976). Optimizing the sensory characteristics and acceptance of cat food is, however, challenging. Cats are sensitive to flavour differences in diet, very discriminative in food selection, and clearly unable to verbalize their likes and dislikes. These issues have dogged the industry for decades.
Behavioural studies with cats are carried out by pet food producers. Simple preference and acceptance tests can be conducted to determine the effects of changes in processing, raw materials, flavourings and presentation. These tests are, however, expensive to maintain, time consuming (Booth, 1976), and yield limited and often equivocal data. In addition, individual animal variation, previous diet or experience, and lateral bias complicate the protocols (Rofe and Anderson, 1970). In conventional flavour research using animals, considerable time and effort is also involved in isolating the required quantities of aroma and fractions, and large amounts of material are required because of concentration-dependent quality changes (Booth, 1976). In light of these limitations and the need for more rapid test procedures, the concept of using human taste panels has been advanced.
While there are differences in the physiological and perceptual systems involved in taste between Felis catus and Homo sapiens, there are also some broad similarities and common sensitivities to stimuli (Boudreau and White, 1978), and evidence suggests that human sensory data can be useful in assisting cat food formulation. Cats use both taste and smell in the detection and selection of food (Bradshaw, 1991). The third chemosensory system, the vomeronasal organ appears to be involved only in the perception of social odours (Hart and Leedy, 1987). Good reviews are available on the sensory capability of cats (Boudreau and White, 1978; Bradshaw, 1991) and on experiential factors that affect feeding behaviour (Bradshaw, 1991). In-house tasting trials using a human taster are commonly conducted by the pet food industry, although there is a paucity of relevant information in the scientific literature. Lin et al. (1998) used a human sensory panel to evaluate the effects of extrusion parameters (fat type, fat content and initial moisture content) on the sensory characteristics of extruded dry pet food during storage. The panel rated the perceived intensities of a set of pre-determined aroma, hue and (manual) texture attributes relative to control samples.
The objectives of this current study were to develop a methodology for using human taste panels to assess canned cat food and to develop base-line flavour profiles for a range of commercial canned cat food products (CFP). The relationship between flavour profiles so-developed and data obtained from cat acceptance/preference trials may enable a more rapid, quantitative and predictive indication of the effects of ingredient and processing changes on the performance of products. Evidence from pet food manufacturers indicate that food flavour, rather than colour or ortho-nasal aroma is dominant in influencing acceptance/preference behaviour in cats. Therefore, this study concentrated on retro-nasal aroma, taste and textural attributes.
Material and methods
Thirteen CFP were used for training and formal testing (Table 1). These consisted of five homogeneous samples and four binary system products, the latter each consisting of a meat chunk (MC) and gravy/gel (GG) component. For binary products, the meat and gravy/gel components were evaluated separately. The manufacturer certified all products sterile and safe for human consumption, and brand names were replaced with alphanumeric codes because of commercial sensitivity. Standardized methods for sample preparation and assessment were developed largely based on panel responses to the screening exercises and discussions during early training sessions. Homogenization of samples to a pâté-like consistency using a domestic blender was trialed in order to minimize sampling variation because of the heterogeneous composition of some products. However, this was discontinued as concerns were expressed that homogenization may have significantly altered the flavour profile of the products. Instead, particular care was taken in the preparation and sampling of each product to ensure that it was representative of the contents of the can Table 1.
|Product code||Product description|
|A||Homogeneous product marketed as minced-beef based|
|B||Homogeneous product marketed as jelly-meat based|
|C1||Product C – binary system: meat chunks portion|
|C2||Product C – binary system: gravy portion|
|F1||Product F – binary system: meat chunks portion|
|F2||Product F – binary system: gravy portion|
|G||Homogeneous product marketed as fish based|
|H1||Product H – binary system: meat chunks portion|
|H2||Product H – binary system: gel portion|
|I1||Product I – binary system: meat chunks portion|
|I2||Product I – binary system: gravy portion|
The sensory methodology used in this study is described below and was generally based on the descriptive analysis techniques used in the human food industry (Lawless and Heymann, 1998).
Prospective panelists were drawn from the Charles Sturt University-Riverina staff and student population with suitability assessed using a questionnaire and tasting exercises. As well as obtaining general background details, including previous panel experience, the questionnaire sought specific information on health-related conditions that might exclude candidates (e.g. sinus problems, colour blindness, food allergies). The screening session comprised a set of four simple exercises in which the sensory acuity of prospective panelists was assessed. In Exercise 1, proficiency in identifying the basic tastes was assessed through blind presentation of four aqueous solutions containing tartaric acid, quinine sulfate, sucrose or sodium chloride. In Exercise 2, proficiency in discriminating different levels of hardness were assessed by having subjects rank the hardness of samples of raw carrot, whole peanuts (unsalted), whole almonds and hard-boiled barley sugar sweets. In Exercise 3, subjects ranked the bitterness of three samples of blended CFP spiked with 0, 0.0128 or 0.0383 mm quinine sulfate. This exercise was repeated once more with a different order of presentation or samples. In the last exercise, subjects tasted three different CFP and rated their hedonic impression using a 9-point balanced hedonic scale, ranging from ‘Like extremely’ to ‘Dislike extremely’ (Peryam and Girardot, 1952). This exercise was considered important as we speculated that subjects with a strong negative attitude towards the products may have reduced motivation, concentration and/or reliability.
The final selection of individuals for the panel was based primarily on satisfactory performance in these exercises and absence of aversion in tasting the products (data not shown). Approximately, 30% of prospective panelists opted not to complete the screening exercise or to continue with subsequent sessions, with dislike of the CFP cited as the most common reason for non-completion. Interest and availability were criteria also used in the selection. The final panel (n = 11) consisted of nine males and two females, with an age range of 19 to 60 years, and comprised six staff and five students from the Charles Sturt University-Riverina campus. Four panelists had previous experience on tasting panels evaluating other food products and seven had no prior experience on tasting panels.
Panel training and protocol development
Descriptor generation, vocabulary and standards development
In total, six 1.5-h training sessions were conducted over a 2-week period during which the panel was introduced to all the products to be formally tested. In the first session, panelists were presented with each of the 13 CFP (Table 1). They were asked to describe the flavour and texture characteristics, as appropriate, using a descriptor generation form. This exercise was repeated without the texture requirement in session 2, generating a combined list of 119 flavour and 25 texture descriptors. In sessions 2–4, the panel re-assessed the samples for flavour along with the proposed reference standards. Discussions of the evaluations were held, and the panel was encouraged to group synonymous terms together where applicable and agree on common descriptors, thus avoiding redundancies (e.g. ‘sour’, ‘tart’, and ‘acid’ were reduced to ‘acid’). Descriptors used by only one panelist were eliminated from the list. A final list of 18 flavour descriptors was derived by panel consensus.
Concurrent with the development of a vocabulary and descriptor list over these sessions was the development of reference standards. The panel was presented with a range of proposed standards for each of the most frequently used flavour descriptors and asked to identify the standard that most closely matched the corresponding sensation elicited by the products. If no match could be found, panelists were encouraged to suggest alternative standards that were then presented in the following session(s). After a suitable reference standard had been developed, its intensity was adjusted (diluted or enhanced) to a level where the panel rated it as being at the upper limit of the range found in the CFP. The approximate order in which the flavour attributes were perceived by the panel was also determined during the training exercises and used in the structuring of the data collection forms. The final lexicon of taste and flavour descriptors and their corresponding reference standards is given in Table 2. Noteworthy are the large number of attributes related to meat flavour, for which amino acids, nucleotides and nitrogen-sulphur compounds are predominantly responsible (Imafidon and Spanier, 1994) and to which cat taste receptors are responsive (Boudreau and White, 1978).
|Attribute||Reference standard composition||Odour(o)/ taste(t)|
|Sweet||0.035 m sucrose in aqueous solution||(t)|
|Sour||4.997 mm tartaric acid in aqueous solution||(t)|
|Tuna||2 teaspoon canned tuna||(o)|
|Herbal||1:1 mix of dried oregano and basil||(o)|
|Spicy||2 teaspoons of Home Brand™ tomato sauce||(o)|
|Soy||3 teaspoons of Home Brand™ soy sauce||(o)|
|Salty||0.0274 m sodium chloride in aqueous solution||(t)|
|Cereal||1:1:1 dry mix of Wheat bix™ + Natura™ Barley Bran + Natura™ Natural Oat Bran||(t)|
|Caramel||one crushed Grannies™ butterscotch||(o)|
|Chicken||boiled and finely diced chicken breast||(o)|
|Methionine||4 teaspoons of methionine powder in 50 ml water; solution boiled then cooled||(o)|
|Vegetable||1 teaspoon of crumbled vegetable stock cube||(o)|
|Offaly||1:1 blend of boiled and finely diced beef kidney and liver – well mixed||(t)|
|Meaty||boiled lean beef mince||(t)|
|Burnt flavour||3:1 aqueous dilution of Reese™ Hickory liquid smoke (4 ml total volume)||(o)|
|Prawn||1 teaspoon xylose powder||(t)|
|Rancid||3 teaspoons of lard (unheated)||(o)|
|Bitter||0.0128 mm quinine sulfate in aqueous solution||(t)|
A similar process was followed in developing a list of appropriate terms for describing the texture of the products. Following evaluation of the frequency of use of descriptors (data not shown) and subsequent panel discussion, it was concluded that the texture of MC was best categorized by the terms hardness, chewiness and grittiness and that of GG by viscosity and grittiness.
Basic tastes recognition and use of scales
In the second training session, panelists were simultaneously presented with coded aqueous reference samples of compounds eliciting the basic tastes (sweet, acid, bitter, salt) and asked to match each of their samples with these tastes. All panelists responded correctly, so this exercise was discontinued in further sessions. During sessions 3 and 4, the panel was introduced to the use of line scales for the intensity rating of the various flavour attributes, including the separated MC and GG components. The exercises typically involved scoring a range of coded products from Table 1 for intensity of each of the flavour attributes on a 15-cm line-scale relative to the appropriate reference standard (which was rated as ‘intense’). After completion of each exercise, the results were collected, assessed and discussed with the panel when instances of inconsistent use of the scales or similar issues were identified.
In session 5, the panel completed exercises similar to those outlined above to familiarize themselves with the use of line scales in rating the texture of both the MC and GG components. Conceptual standards (‘reference foods’) were advanced and modified by the panel to represent the extremes of intensity for each of these textural terms. For MC, the scale anchor terms and conceptual standards were; hardness: ‘very soft’ (loosely-set jelly), ‘hard’ (raw carrot); chewiness: ‘very tender’ (sushi/raw fish), ‘very tough’ (over-cooked steak or fruit leather); grittiness: ‘not gritty’ (set jelly), ‘very gritty’ (fish-bone portion of canned fish). For GG, the scale anchor terms and conceptual standards for viscosity were ‘very thin’ (tap water) and ‘very thick’ (condensed milk), while those used for grittiness were the same as for MC. In session 6, a ‘dummy run’ was held of the formal profiling/data collection sessions to follow in order to familiarize the panel with the tasks required.
During the training sessions, standardized evaluation protocols for the products were developed and refined with extensive input from the panel. The final tasting protocol for assessing flavour is described below. Prior to entering the tasting booths, panelists familiarized themselves with the reference samples. Each sample was then evaluated as follows: (i) mouth rinse with water; (ii) 0.5–1 teaspoon of sample taken onto a teaspoon and placed in mouth; (iii) sample moved around mouth and chewed for 10–15 s; (iv) a portion of the sample swallowed and the remainder expectorated into a spittoon; (v) intensity ratings for each attribute made on 15-cm line-scale; (vi) mouth rinse with water. A 1–2 min break was enforced between samples.
The final tasting protocol for assessing texture is described below. Meat Chunk was assessed for hardness, chewiness, and grittiness as follows: (i) mouth rinse with water; (ii) 1–2 MCs taken onto a teaspoon and placed in mouth; (iii) sample chewed using molars until masticated to the point of being ready to swallow; (iv) sample expectorated; (v) intensity ratings for each attribute made on 15-cm line-scale; (vi) mouth rinse with water followed by a 1–2 min break between samples. Gravy/gel was assessed for viscosity and grittiness as follows: (i) mouth rinse with water; (ii) 0.5–1 teaspoon of gel/gravy taken onto a teaspoon; (iii) teaspoon placed close to lips and air drawn in gently to induce flow of the liquid (panelist attends to the force required to induce flow); (iv) sample placed in mouth and moved across tongue (panelist attends to the perceived rate of flow and grittiness); (v) sample expectorated and intensity ratings made on 15 cm line-scale; (vi) mouth rinse with water followed by a 1–2 min break between samples.
Testing and data analysis
Formal profiling of the CFP took place over three consecutive sessions and days at the sensory evaluation laboratory of the National Wine and Grape Industry Centre, Charles Sturt University-Riverina. At the commencement of each session, panelists were reminded through verbal instruction of the correct protocols for flavour and/or texture assessments, and this information was repeated on the forms used to record the panel responses. Emphasis was also placed on techniques for avoiding sensory fatigue and carry-over effects during the testing (e.g. taking time, resting between samples, use of water). The 13 commercial CFP (Table 1) were assessed over two sessions for the 18 flavour attributes (Table 2) in duplicate using a complete block design, and the order of presentation of samples randomized. Six to seven coded samples/flight were presented simultaneously (two flights/sessions) at room temperature in 62 ml plastic sample cups. The protocols described earlier were used and reference standards were also available throughout the testing period as required.
The flavour intensity of each attribute was indicated by scoring with a small vertical line on a 15-cm scale anchored by the terms ‘absent’ and ‘intense’, with ‘intense’ corresponding to the appropriate reference standard. In addition, liking was measured for each product using a 9-point hedonic scale (Peryam and Girardot, 1952). A 10-min minimum break was forced between the two flights each day.
The components of the binary samples (C1, C2, F1, F2, H1, H2, I1 and I2) were assessed for texture in duplicate on the final day of testing. Each product was separated into its respective MC and GG component and presented as a coded sample at room temperature in 62 ml plastic sample cups, and the order of presentation randomized. All samples were presented in a flight and with a 10-min break enforced between the two flights. The intensity of attributes was measured using 15-cm line-scales anchored with the terms and conceptual reference foods described above. For both flavour and texture, assessment panelists were also encouraged to add any additional descriptors in the appropriate section of their evaluation forms. For all attribute data, panel marks were converted into a score out of 15 (distance on the scale in cm from the first anchor point). xlstat© version 7.5.2 (Addinsoft, Paris, France) was used for all statistical analyses. anova was conducted with panelist, product and replicate terms in the model, and Tukey’s honesty significant difference (HSD).05 was used as the means separation test. Principal components analysis (without rotation; PCA) was completed for the flavour data, and Pearson’s correlation (PC) coefficients calculated for both flavour and texture data.
Results and Discussion
anova showed that CFP differed in mean intensity ratings for 16 of the 18 flavour attributes. Specifically; sweet (p < 0.0001), sour/acid (p < 0.0001), tuna (p < 0.0001), spicy (p = 0.000), soy (p < 0.0001), salty (p < 0.0001), cereal (p < 0.0001), caramel (p < 0.0001), chicken (p < 0.0001), methionine (p = 0.000), offaly (p < 0.0001), meaty (p < 0.0001), burnt flavour (p < 0.0001), prawn (p < 0.0001), rancid (p < 0.0001) and bitter (p < 0.0001). Averaged across all products, salty (5.39), sweet (5.28), and meaty (5.03) scored highest for intensity by the panel, while the lowest attribute scores were for tuna (1.48), vegetable (1.83), prawn (1.90) and chicken (1.98).
Meat Chunks differed in intensity to GG components for nine flavour attributes. The average intensity scores of those attributes are, for MC and GG respectively: sweet, 4.6 and 7.1 (p < 0.0001); herbal, 1.8 and 2.6 (p = 0.017); spicy, 2.1 and 3.0 (p = 0.011); salty, 4.2 and 5.5 (p < 0.0001); cereal, 5.9 and 3.2 (p < 0.0001); caramel, 1.8 and 3.3 (p < 0.0001); meaty, 6.0 and 4.3 (p < 0.0001); rancid, 3.7 and 2.9 (p = 0.025) and bitter, 3.6 and 2.3 (p = 0.001). Cereal and meaty were the attributes rated most intense for 3 of the 4 MC products (C1, H1, I1), while sweet and meat were the most intense attributes for F1 (data not shown). Overall, panelists were a significant (p < 0.05) source of variation for all attributes, while no replicate differences were found for any attribute (p > 0.05). Variation between panelists in their intensity scores is commonly reported in sensory studies, and may be attributable to individual differences in sensory acuity and/or idiosyncratic use of the line scales.
Table 3 shows the correlation coefficients for the flavour terms. All descriptors show significant associations with other descriptors, although they are generally weak (−0.5 < x < 0.5). The exception is rancid, which has a moderate positive correlation with offaly (0.705), perhaps indicative of a common stimuli (e.g. offal) in the products. Some conceptually similar constructs are positively correlated, such as sweet/caramel (0.394), burnt/bitter (0.406) and salty/soy (0.204), while the negative association between tuna and most other terms may reflect the dominating influence of this attribute when fish components are included in product formulations. Methionine, an uncommon descriptor in human food profiles is most strongly correlated with rancid (0.366), vegetable (0.326) and bitter (0.323) and is a reported additive in pet food (personal communication)Table 3.
Figure 1 displays the first two components of the PCA, which account for 52.8% of the variation in the data. PC1 is positively loaded with salty and prawn, and negatively loaded with meaty, offal and burnt flavour, and thus may be construed as a ‘fish’ vs. ‘red meat’ axis. PC2 is largely defined as a sweet/caramel/herbal vs. acid axis, although chicken is negatively loaded here as well. There is relatively little separation between the products in this PC space with the exception of G, which is separated from the rest based on higher scores for prawn and tuna attributes. This product was indeed, labelled as fish-based food. F2 is somewhat separated from other products, based on its higher intensity scores for sweetness and caramel. Figure 2 displays the third and fourth components of the PCA, which account for 30% of the variation in the data. PC3 is positively loaded with acid and spicy, and negatively loaded with cereal, while PC4 is largely defined by its strong positive loading for chicken. There is generally very good separation of products within this PC space. Some general groupings are also suggested: (i) both the MC and GG portions of F, which have similar scores for sweet, caramel and rancid, (ii) G, A and I2, which have similar ratings for salt intensity, and (iii) H1 and C2. Overall, product pairs A–I2, and C2–H1 appear most similar to each other in flavour profile, based on their relative proximity in both PCA spaces.
Texture and hedonic ratings
Significant differences in scores for the four MC components were apparent for all three texture attributes assessed (Figure 3). F1 had lower scores than other products for hardness, chewiness and grittiness, while I1 had higher scores for ‘chewiness’ and ‘hardness’. C1 had the highest rating for ‘grittiness’ of all products. Grittiness appeared to be the attribute that discriminates most between these products, and had the greatest intensity score range (5.7). Pearson’s correlation coefficients were calculated as follows – hardness:chewiness (0.695); hardness:grittiness (0.416); chewiness:grittiness (0.469). All associations were significant at p < 0.05. All associations were significant at p < 0.05. The hardness:chewiness association was expected, as chewiness is a composite term that includes hardness (ASTM International, 2003). Significant differences between the four GG components were also observed for both viscosity and grittiness (Fig. 4). In particular, H2 scored highest for viscosity, and C2 for grittiness. Overall, grittiness ratings for all products were very low, suggesting it may have limited utility for future profiling of cat food. Pearson’s correlation coefficient for grittiness:viscosity was −0.039 and was not significant (p > 0.05). Panelists were a significant (p < 0.05) source of variation for all texture attributes for both MC and GG, while no replicate differences were found (p > 0.05) Fig. 4.
Given the anecdotal observation that cats sometimes display preference for one or another of the components of binary systems, and that this panel has found many differences in flavour and texture between MC and GG, we recommend that these components continue to be assessed separately in future sensory research with CFP. A more detailed study examining CFP texture may be appropriate, should this domain prove particularly important for cat feeding behaviour. This may include further consideration of adhesiveness and particle size, shape, and orientation.
Additionally, determination of the time-intensity characteristics of the key flavour and texture attributes identified here may yield information relevant to predicting food acceptance/preference behaviour in cats, as would consideration of ortho-nasal cues.
For hedonic responses, the data was coded one to nine for analysis, corresponding to ‘like extremely’ to ‘dislike extremely’ respectively. anova showed that products (p < 0.0001) and panelists (p < 0.0001) were significant sources of variation, while hedonic scores did not vary across replicate sessions (p = 0.475). Product G was most liked (mean = 2.73), while E was least liked (6.59). No difference in liking was found between the three product types (homogenous, MC, GG) (p = 0.353). Perhaps surprisingly, the grand mean of all hedonic scores was 4.97, placing it between the ‘neither like nor dislike’ and ‘like slightly’ scale adjectives. Liking score was positively correlated with rancid (0.337, p < 0.0001), offaly (0.291, p < 0.0001), cereal (0.193, p = 0.001), burnt flavour (0.185, p = 0.002), methionine (0.168, p = 0.004) and bitter (0.166, p = 0.005), meaning actual liking decreased as these attributes increased in intensity. Conversely, it was negatively correlated with tuna (−0.271, p < 0.0001) and herbal (−0.231, p < 0.0001) scores.
The author gratefully acknowledge the support of the industrial partner on this study. Shane Walsh, Donella Geddes and Amy Blake are sincerely thanked for their advice and technical assistance.
- ASTM International, 2003: E 1909–97 (Reapproved 2003). Standard Guide for Time-Intensity Evaluation of Sensory Attributes. ASTM International, West Conshohocken, PA, USA.
- 1976: Why flavour research in the pet food industry? International Flavours and Food Additives 7, 11–12, 16. ,
- 1978: Flavor chemistry of carnivore taste systems. In: R. W.Bullard (ed.), Flavor Chemistry of Animal Foods, American Chemical Society, Washington, DC, USA, pp. 102–128. ; ,
- 1991: Sensory and experiential factors in the design of foods for domestic dogs and cats. The Proceedings of the Nutrition Society 50, 99–106. ,
- 1987: Stimulus and hormonal determinants of flehmen behaviour in cats. Hormones and Behavior 21, 44–52. ; ,
- 1994: Unraveling the secret of meat flavor. Trends in Food Science & Technology 5, 315–321. ; ,
- 1998: Sensory Evaluation of Food: principles and practices. Chapman and Hall, New York. ; ,
- 2006: Cats lack a sweet taste receptor. The Journal of Nutrition 136, 1932S–1934S. ; ; ; ; ; ; ; ; ; ; ,
- 1998: Effects of lipids and processing conditions on the sensory characteristics of extruded dry pet food. Journal of Food Quality 21, 265–284. ; ; ; ,
- 1996: Multivariate Analysis of Data in Sensory Science. Elsevier Science, Amsterdam. ; ,
- 1952: Advanced taste test method. Food Engineering 24, 58–61, 194. ;
- 1970: Food preference in domestic pets. The Proceedings of the Nutrition Society 29, 330–335. ; ,