The research on puppy testing is not the blank slate it is sometimes described as, nor the endorsement the test charts imply — it is a reasonably consistent picture in which single sessions at seven or eight weeks predict adult behavior poorly, later and repeated assessment predicts distinctly better, and the quality of the test as an instrument is a separate question from whether it predicts anything at all.
The most important first
- Reliability and validity are different problems: a test can produce the same score twice and still predict nothing, and a systematic review of working-dog behavioral tests found that of 16 qualifying papers only three or four reported inter-rater reliability and four test-retest reliability.
- Behavior in puppies is measurably less stable than in adults. A meta-analysis of 31 studies found an overall consistency of r = 0.43, rising to r = 0.51 in adult dogs but falling to r = 0.30 in puppies.
- Consistency varies by trait. In puppies, aggression (r = 0.51) and submissiveness (r = 0.43) held up best; responsiveness to training (r = 0.16) and fearfulness (r = 0.24) held up worst.
- The largest single-breed test of an eight-week protocol — 630 German shepherd puppies, retested at 450 to 600 days — found correspondence to adult performance negligible, and the test not useful for predicting suitability for service work.
- Later testing does better: in 93 guide dog puppies assessed at around five and eight months, seven measures at five months and five at eight months were significantly associated with qualifying or being withdrawn from training.
Reliability and validity are two separate questions
Most disagreements about puppy testing are confusions between these two ideas. Reliability asks whether the test measures consistently — whether two observers score the same puppy alike, whether one observer scores the same video alike twice, and whether the puppy scores similarly on a second run. Validity asks whether the score corresponds to anything you care about: here, whether it predicts the adult dog. A test can be highly reliable and entirely invalid.
The systematic review by Brady and colleagues in Frontiers in Veterinary Science (2018) screened 13,691 records and found 16 papers meeting its criteria for behavioral tests assessing characteristics important in working dogs. Its central finding was an absence rather than a result: a widespread lack of reported information on reliability and validity, with inconsistent terminology and definitions of success that made studies hard to compare. When only a quarter of published tests report whether two assessors agree, the literature cannot yet say how good the better tests are.
What the consistency literature shows
The most useful single summary is the meta-analysis by Fratkin and colleagues in PLOS ONE (2013), which pooled 31 studies of temporal consistency in dog behavior. Across all of them consistency was moderate: r = 0.43 between a dog’s earlier and later measurement. That is substantial — dogs are not random — but it leaves most of the variance in later behavior unexplained by the earlier score.
The breakdown matters more than the headline. Adult dogs were markedly more consistent than puppies, r = 0.51 against r = 0.30. Within puppies, some traits traveled better than others: aggression at r = 0.51 and submissiveness at r = 0.43, against responsiveness to training at r = 0.16 and fearfulness at r = 0.24. Sociability gave inconsistent results; in adults the dimensions behaved much alike.
Two methodological findings should change how you read any result. In adult dogs, consistency fell as the interval lengthened — r = 0.60 under ten weeks against r = 0.32 beyond twenty-four. Across the whole meta-analysis, though, the test interval explained only about 2% of the variation, and puppy estimates were similar whatever the interval. And using the identical method at both time points produced roughly 1.8 times the consistency of using different methods. Much of what looks like an unstable dog is an unstable measurement.
One result cuts against the obvious interpretation. Consistency was no lower for dogs tested first as puppies and then as adults than for dogs tested twice as puppies. That locates the problem in puppyhood itself rather than in some barrier between puppyhood and adulthood: a low score today may not even predict next month’s score.
The negative case, properly stated
The most direct test of a breeder-style protocol remains Wilsson and Sundgren’s study in Applied Animal Behaviour Science (1998), which put 630 German shepherd puppies through a ten-part behavior test at eight weeks and followed them to a standardized assessment at 450 to 600 days. The tested traits showed medium-high to high heritability, so the test was measuring something real and partly genetic. But correspondence to adult performance was negligible, and the authors concluded the puppy test was not useful for predicting adult suitability for service dog work. They also found maternal effects on the puppy scores, noting such effects are more likely in juvenile than adult behavior — part of what an eight-week score registers is the litter and the dam, not the individual.
Note the shape of that conclusion: heritable, measurable, reliable enough to compute heritabilities on, and still not predictive.
The pet-dog evidence points the same way. In a Border collie study that followed puppies from their first days to adulthood, only one of ten traits measured at six to seven weeks showed a clear link to behavior at one and a half to two years, and the authors called early tests’ predictive value limited.
Where prediction does work
The more optimistic evidence comes from organizations that follow every dog to an outcome, and it points to two things: test later, and measure the right constructs.
Harvey and colleagues, in the Journal of Veterinary Behavior (2016), tested 93 guide dog puppies at a mean of 4.8 and 8.0 months. Consistency between the two sessions was good for some measures — jumping up, barking, a low greeting posture — and poor for others, including obedience responses, lip licking, body shaking and scratching: within one test, some items behave like traits and others like moods. Seven measures at five months and five at eight months were significantly associated with whether the dog later qualified or was withdrawn, and logistic regression models built from them showed real promise as a tool for identifying unsuitable dogs early in training.
The systematic review adds a pattern across programs: positive-affect measures such as willingness to work and social behavior predicted guide dog and police or military outcomes more strongly, while negative-affect measures such as aggression and sensitivity to aversives predicted inconsistently. It also makes the fairest point available on the negative results — a lack of predictive validity may reflect the dog’s development at the age it was tested rather than a weakness of the test.
The programs that do predict well test later, test more than once, and track every dog’s outcome. A US detection-dog program found that tests from about three months onward predicted which dogs were selected, and that how calmly a dog handled new environments mattered at every age. Even so, a study of several thousand assistance dogs assessed at about one year and eighteen months reached only around two-thirds accuracy. The models were better at spotting dogs likely to fail than at picking dogs likely to succeed. A single test at seven weeks is working with much less.
In practice
A breeder with a working line has tested every litter at seven weeks for a decade and kept the sheets. Reading the literature, the breeder does something uncomfortable: pulls the records for thirty dogs now aged three or more and writes down what became of each. One column lines up loosely — the puppies scored most reactive to the noise item are over-represented among the dogs whose owners later report noise problems. The rest, including the trainability scores the breeder leaned on hardest when placing, show nothing. The seven-week test stays, with two items; the decisions move to a second assessment at five months.
(Illustrative, composite scenario.)
What this justifies concluding
Three statements are supportable. A single session at seven or eight weeks is a poor predictor of adult behavior, and the confident claims attached to the classic charts are not backed by follow-up data. Assessment improves with age and repetition — the same instrument used twice, months apart, consistently administered by someone who did not raise the litter. And prediction is trait-specific rather than global: modest traction on something like aggression or a specific sensitivity, almost none on trainability.
Three limitations belong with them. Most of this work is on single breeds, often German shepherds, Labradors and Border collies, in working or guide dog populations selected for the purpose, so it may generalize poorly. The outcomes in the strongest studies are program-specific — qualifying as a guide dog is not the same as being a settled family pet. And almost none of it tests what breeders actually do, which is not a formal protocol but eight weeks of accumulated observation; that is an absence of studies, not a finding against it.
It makes early assessment a source of hypotheses rather than conclusions — the position puppy aptitude tests and their limits sets out in practical terms, and the reason observing and documenting individual puppies carries more weight here than any scoring sheet.
A test result also cannot diagnose a problem. A puppy that is markedly fearful or reacts aggressively needs a veterinary assessment, not just a lower score.
Work on it
A few minutes here will teach you more than re-reading the article.
Ask yourself
- Which specific claim am I making when I describe a puppy to a buyer — and would the evidence support it for that trait, at that age?
- Do I administer my assessment the same way every time, or has it drifted between litters so that comparison is meaningless?
- If I tested at five months instead of seven weeks, what would I actually have to change to make that possible?
Find out & check
- Go back through your own records for dogs now over two years old, and check your early notes against what their owners report today — one trait at a time, not overall.
Do one thing
- Pick the single trait you most often predict at eight weeks, and write down how you would know in three years whether you were right.
Was this article helpful?
Anonymous. We do not store your name, email or IP address.
Sources & further reading
- Fratkin, J. L. et al., PLOS ONE 2013;8(1):e54907 — Personality Consistency in Dogs: A Meta-Analysis. journals.plos.org
- Brady, K., Cracknell, N., Zulch, H. & Mills, D. S., Frontiers in Veterinary Science 2018;5:103 — A Systematic Review of the Reliability and Validity of Behavioural Tests Used to Assess Behavioural Characteristics Important in Working Dogs. frontiersin.org
- Wilsson, E. & Sundgren, P.-E., Applied Animal Behaviour Science 1998;58(1–2):151–162 — Behaviour test for eight-week old puppies—heritabilities of tested behaviour traits and its correspondence to later behaviour. sciencedirect.com
- Harvey, N. D. et al., Journal of Veterinary Behavior 2016;11:65–76 — Test-retest reliability and predictive validity of a juvenile guide dog behavior test. sciencedirect.com
- Riemer, S. et al., PLOS ONE 2014;9(7):e101237 — The Predictive Value of Early Behavioural Assessments in Pet Dogs – A Longitudinal Study from Neonates to Adults. journals.plos.org
- Bray, E. E. et al., Frontiers in Veterinary Science 2019;6:49 — Predictive Models of Assistance Dog Training Outcomes Using the Canine Behavioral Assessment and Research Questionnaire and a Standardized Temperament Evaluation. frontiersin.org
- Lazarowski, L. et al., Animals 2021;11(4):993 — Validation of a Behavior Test for Predicting Puppies’ Suitability as Detection Dogs. pmc.ncbi.nlm.nih.gov
- Alberghina, D. et al., PLOS ONE 2020;15(7):e0236271 — Behavior test for seven-week old puppies. journals.plos.org
Where evidence is limited, recommendations rest on established practice and expert judgment, and may be updated as knowledge develops. Correlation coefficients describe populations, not individual dogs, and most of the studies above were conducted on single breeds in working or assistance dog programs.
Next steps
- Read next: Evaluating conformation and physical development in puppies.
- Related: Puppy aptitude tests and their limits.
- Return to the area hub: Understanding and placing each puppy.