For a comprehensive review of the visual breed identification literature, please see National Canine Research Council's complete analysis here

Breed labels assigned to dogs of unknown origin are often inaccurate, and observers don't agree on visual breed identification.

 
Research has consistently shown that visual breed identification is very often inaccurate1,2,3,4, and scientists have known for decades that even first generation crossbreeds usually look dramatically different than either parent5. Moreover, recent research also indicates that even experts have very little agreement when visually determining breed2,3,4,6.

In the video below, Dr. Voith describes her initial 2009 research1 which compared the adoption agency visual breed identifications of 20 mixed-breed dogs against DNA identification. The results showed little agreement between reported breed (from visual identification) and actual breed as determined by DNA analysis. Of the 16 dogs that had been assigned a specific breed by their adopting agency, DNA analysis showed that only 4 were actually predominantly comprised of said breed. Furthermore, for 3 of the 4 that did have genetic markers matching their visually identified breed, that breed contributed at most 12.5% of their makeup. Finally, those 3 dogs also had other breeds of equal contribution that were not identified by their adopting agencies. In all only 31% of the dogs showed any DNA evidence of the adoption agencies’ named breeds somewhere in the dogs’ ancestry. Over 90% of the dogs did not have their visually identified breeds as the predominant breed in their DNA analysis.

For additional information, please see:

 
A second study2 by Dr. Voith and colleagues looked at both the accuracy of visual breed identification and inter-observer reliability, even among qualified canine experts, using the same dogs as the above-referenced study. Both were very low. Fewer than half of the 923 participants were able to correctly identify the predominant breed (based on DNA analysis) for 14 of the 20 dogs. Moreover, for 3 of the dogs visual identification did not match any (major or minor) DNA breed identification. More than 70% of the study participants reported that at the time of the study, or at one time, their breed descriptors were used in record keeping, yet the observers were unable to agree with each other or with the DNA identification. For only 7 dogs (35%) could even 1/2 the observers agree on a predominant breed.
 

For additional information, please see:

  • Poster Version of the Study:

 
Separate, additional visual breed identification research was conducted at the College of Veterinary Medicine at the University of Florida3,4, which further confirmed the unreliability of visual breed identification used by dog adoption agencies, animal control, and in regulation.
 
The first3 was an expanded 2012 survey of over 5,000 dog experts - veterinarians, breeders, trainers, shelter staff, groomers, rescuers, and others - who visually assessed breeds in dogs in a series of photographs. Their assessments were then compared to DNA breed profiles of the dogs. For the purposes of the survey, a response was considered "accurate" if it named any of the breeds DNA analysis had detected in the dog, no matter how many other breeds had been detected, and whether or not the breed guessed was a predominant breed in the dog. Since, in almost every dog multiple breeds had been detected, there were lots of opportunities to be correct. However, the respondents were only correct in naming at least one of the breeds detected by DNA analysis less than 1/3 of the time. And no profession did significantly better than any other. Every profession’s responses, in total, were correct less than 1/3 of the time.
 

For additional information, please see:

  • How long before we discard visual breed identification?

 
The second study4, which was published in 2015, also explored reliability of visual breed identification when compared to DNA, as well as consistency among experts. The researchers specifically focused on visual identification of “pit bull-type dogs,” because they are commonly targeted in breed-specific legislation (BSL). Because the term “pit bull” does not denote a specific breed and there is no agreed-upon definition of “pit bull-type dogs,” for the study the authors defined the term to include Staffordshire bull terriers, American Staffordshire terriers, and American pit bull terriers. The 16 study participants were specifically selected because of their routine visual breed-identification duties, hence the decisions made by this group regularly have real-world consequences. Participants visually identified dogs in their kennels, and then assigned primary and secondary breed labels to each dog. If a participant labeled a dog as any of the aforementioned breeds or “pit bull” or “mix” their breed assignment was considered a DNA match for the study if the test results showed 12.5% or more of American Staffordshire Terrier or Staffordshire Bull Terrier. Thus the criteria for a “match” were extremely broad, including multiple breeds and very low DNA material percentage.
 
The median agreement between visual identification and DNA analysis for discriminating between “pit bull-type dogs” and “non pit bull-type dogs” ranged from 67-78%. The average sensitivity for identifying “pit bull-type dogs” was 50%. Of the 95 dogs who fell into the “not pit bull-type” group by DNA analysis, 38% were nevertheless given the label by at least one breed assessor. That is a significant finding; of the dogs whose DNA did not reveal contributions from Staffordshire bull terrier or American Staffordshire terrier, more than 1 out of 3 were still labeled “pit bull.” In sum, the overall validity (specificity and sensitivity results) even with such a broad target for identification, did not reach the level of a coin toss.
 

For additional information, please see:

 
An additional study6 published in 2014 sought to compare U.S. and U.K shelter staff and volunteer’s inter-observer reliability, in particular, determining the level of agreement between the two groups regarding dogs they labeled “pit bulls.” There were 470 participants who indicated that their role involved assigning breed to dogs based on visual identification. Participants completed a breed identification survey online in which they viewed 20 dogs and completed questions such as what breed they would assign to each dog and whether they consider the dog to be a “pit bull.” There was not strong agreement between or within the shelter workers in the two countries in terms of breed identification, and like Olson, et al.4, this study also showed that there were discrepancies regarding which breeds are considered to be “pit bulls” (a term which has no single legal, municipal, or standard definition). Unlike the other studies1,2,3,4, there was no attempt to establish either group’s accuracy (e.g., by comparison with DNA results).
 

For additional information, please see:

​

 

Implications for Research & Policy

 
It has been customary in our society to look at a dog and guess its breed or breed composition. Many scenarios result in a dog’s presumed breed being recorded, including arrival at a shelter, veterinary visits, licensing, housing and insurance applications, doctor visits following dog bites, and the subsequent media reporting of said incidents. Eventually these unconfirmed identifications make their way into databases that are then used in retrospective research studies to make claims about canine behavior. Several studies have sought to relate breed and behavior, dog bites and dog bite-related fatalities. These studies influence public opinion and policy regarding particular breeds. Inaccurate data regarding breed are being propagated, studied, and used to drive policy.
 
The above visual breed identification studies’ results bring into question the findings of any study which attempts to link breed to behavior based on visually identified study populations and demonstrate a need for eliminating visual breed identification as a data source for ongoing canine behavioral studies.
 
An article published in 20127 considered the implications of these undisputed findings for veterinary practice, and recommended that veterinarians stop attempting to assign breed labels to mixed-breed dogs whose parentage they do not know.
 
Visual breed identification is inconsistent and unreliable, and the research surrounding it should deter scientists, media outlets, veterinarians, shelter staff, and policy makers from relying on visual breed identification, as well as making broad statements about breed and behavior, absent pedigree documentation.
 

For additional information, please see:

​

 
 

Reliability of DNA Breed Identification

We take very seriously the reliability of the studies on which we report and understand that there are those who are skeptical of breed identification obtained through DNA analysis. Indeed, it is important to note that DNA identification may not be 100% accurate when analyzing mixed-breed dogs, nor do the companies which conduct the analyses claim it to be so. At the time Dr. Voith conducted the first of these studies1, the accuracy of the Mars Wisdom Panel® used in the studies was reported to be 84%, for identification of breed in F1 crosses (offspring of 2 different registered purebreds). Accuracy is not currently reported in a percentage, however Mars Wisdom Panel®, which is specifically intended for mixed-breed dogs, was developed by analyzing more than 19 million genetic markers taken from 13,000 dogs.
The results of the above-mentioned studies allow us to say with confidence that the accuracy for DNA analysis is much higher than that achieved by looking at the dog for at least 2 reasons:
  • Dr Voith’s second study2 determined that the agreement as to breed designation among people in dog-related professions was extremely low, with less than 1/3 of the people agreeing on a predominant breed for more than 1/2 the dogs evaluated. Olson et al.4 and the second survey conducted at the University of Florida3 also showed poor agreement among observers, in addition to the poor correlation with breed identification by DNA analysis.
  • In Genetics and the Social Behavior of the Dog5, a seminal work on dogs and the significance of documented pedigree, the authors showed that that even F1 crosses very rarely have much physical resemblance to either of their parents’ breeds.  
Mated pair -- basenji & cocker BCS F1 hybrids -- male and female pair
Scott Fuller; Mated pair: basenji, cocker

Scott, J.P., & Fuller J.L. (1965). Genetics and the Social Behavior of the Dog. Chicago, IL: University of Chicago Press.  

Understanding how a dog's appearance is determined by its DNA can also help explain why a DNA test is better than visual breed identification. Visual identification is based upon the observation of a handful of variable breed-associated physical traits, such as coat color, body size, skull shape and whether the ears or erect or floppy. These physical traits are found in many different breeds and are controlled by approximately 50 of the roughly 20,000 genes that create a dog. Sometimes, a breed may exhibit a certain physical trait because all the members in the breed have exactly the same version of the gene that encodes the trait.
 
If this trait is recessive (for example like the trait associated with long fur), only dogs with two of the same version of the gene will exhibit long fur. If one of these dogs is the ancestor of a mixed- breed dog, the mixed-breed dog may contain both the DNA for the recessive version of the trait (long fur) and the dominant version of the trait (short fur). However, the long-haired recessive appearance will not be observed because the dominant short-haired DNA would determine the visual appearance of coat length (making it short). Subsequently, the visual identification of breed would inaccurately specify short-haired breeds based upon the visual observation of short hair.
 
The DNA test would be able to detect the recessive version of the gene for the long hair along with the dominant version of DNA for the short-hair and the DNA test result would use that information to determine the breed. The DNA results might report both long-haired and short-haired breeds in the dog's ancestry even though the dog only has short-hair. Coat length is not the only trait that can be "hidden" from visual observation due to dominant and recessive patterns of genetic inheritance in dogs.
 
Although the genetic test may not assess every gene or even each physical attribute of a dog, the regions of the genome that it uses to assess breed take into account much more information than visual observation. The DNA test is better than visual breed identification because it takes into account the pattern of genetic variation at many different regions across the dog genome to generate a "genetic snapshot" of a mixed-breed dog's ancestry. The resulting genetic evidence for what breeds make up a mixed-breed dog may or may not agree with visual observations, but they do agree with what scientists have discovered from two decades of sequencing and studying genomes.
 
While breed identification by DNA analysis is more accurate than visual breed identification, it’s important to remember that neither identifies genetic markers influencing specific traits or predicts behavior of any particular dog. Each dog is an individual, and its physical and behavioral traits will be the result of multiple factors.
 

For a graphic illustration of these concepts, please see the Infographic by clicking here.

 

Expand your understanding with these additional resources: 

       Updated July 13, 2016

U           

       SOURCES and NOTES:

  1. Voith, V. L., Ingram, E., Mitsouras, K., & Irizarry, K. (2009). Comparison of Adoption Agency Breed Identification and DNA Breed Identification of Dogs. Journal of Applied Animal Welfare Science, 12(3), 253-262.
  2. Voith, V. L., Trevejo, R., Dowling-Guyer, S., Chadik, C., Marder, A., Johnson, V., & Irizarry, K. (2013). Comparison of visual and DNA breed identification of dogs and inter-observer reliability, American Journal of Sociological Research, 3(2) 17-29.
  3. Croy, K. et al. (2012). “Dog Breed Identification: What kind of dog is that?” Retrieved from: http://sheltermedicine.vetmed.ufl.edu/library/research-studies/current-s...
  4. Olson, K. R., Levy, J. K., Norby, B., Crandall, M. M., Broadhurst, J. E., Jacks, S., Barton, R. C., & Zimmerman, M. S. (2015). Inconsistent identification of pit bull-type dogs by shelter staff. The Veterinary Journal, 206, 197-202.
  5. Scott, J.P., & Fuller J.L. (1965). Genetics and the Social Behavior of the Dog. Chicago, IL: University of Chicago Press. 
  6. Hoffman, C. L., Harrison, N., Wolff, L., & Westgarth, C. (2014). Is that dog a pit bull? A cross-country comparison of perceptions of shelter workers regarding breed identification.  Journal of Applied Animal Welfare Science: JAAWS, 17 (4), 322-339.
  7. Simpson, R. J., Simpson, K., & VanKavage, L. (2012). Rethinking dog breed identification in veterinary practice. Journal of the American Veterinary Medical Association, 241(9), 1163-1166.