Temperament Testing a Service Dog Candidate: What Actually Predicts Success

Updated September 10, 2026 · first published December 14, 2021Rewritten in full against the current regulations

Last updated: 27 August 2026

Somebody is going to offer to temperament test a puppy for you, probably at seven or eight weeks old, probably for a fee, and probably at the exact moment you are deciding whether to hand over money for a dog.

The research on whether that test tells you anything is a lot clearer than the people selling it let on.

It tells you something. It does not tell you nearly enough to make the decision on, and it fails in a specific direction that almost nobody explains.

Start with the number that should end the seven-week puppy test

Researchers followed 465 puppies through a standardized behavior test and then through guide dog training, to see whether the test predicted which ones would qualify.

Of the 450 puppies that scored above the proposed cut-off, 66 percent went on to certify.

In the sample as a whole — including the ones the test would have rejected — 64 percent certified.

Passing the test moved the odds by two percentage points. The finding is described in an open-access 2014 paper in PLoS ONE, which reports it as “low predictability of successful certification.”

That is not a broken test. That is a test doing roughly what a coin does.

Why early tests fail, in one sentence

The largest analysis of this question is a meta-analysis in PLoS ONE covering 31 studies and 822 effect-size estimates. It found that dog personality is genuinely consistent overall, at r = 0.43 — so this is not a paper arguing that temperament is meaningless.

But look at where the consistency sits:

An r of 0.30 means the puppy measurement accounts for roughly nine percent of the variation in the later one. Real, but nowhere near a selection decision.

And here is the part that matters

The same meta-analysis broke the puppy figure down by trait:

The puppy test is most reliable at measuring the one trait that only rules a dog out, and least reliable at measuring the three traits the job actually requires.

That is the whole problem, and it is fully documented. The authors are careful about it too — their own line is that “the blanket idea that ‘puppy tests do not work’ needs to be reconsidered”, because it depends which dimension you are asking about. Fair. But if the dimension you care about is trainability, the answer is 0.16.

The finding that should change how you read any test result

This is the most useful thing in the entire literature and it is almost never mentioned.

A 2025 study in Frontiers in Veterinary Science analyzed 15,374 behavior evaluations across 3,969 Labrador Retrievers in a guide dog program, with up to eight assessments per dog from seven weeks through final training.

On the seven-to-eight-week puppy test, the authors are blunt:

“Generally, the P does not agree with any other observation time point.”

But the finding to carry with you is this. The tests were highly accurate when they said yes, and close to useless when they said no.

In the authors’ words: “Dogs who score well early in life are likely to consistently score well and the dog’s behavior is unlikely to regress over time” — while “the predictive capacity of the BCL is limited if a dog scores poorly.”

A good result is trustworthy. A bad result is close to meaningless. A dog that fails a temperament assessment may well go on to succeed.

Now notice what every commercial temperament test is sold to do. Screen out. Find the bad ones. Tell you which puppy not to buy.

That is the exact direction the evidence does not support.

One honest caveat on that study

The authors flag it themselves: between “15 to 30% of puppies released” each year were dropped from the dataset because they never got a second evaluation. Removing dogs that failed early and left makes the remaining picture look more stable than it is. It does not overturn the asymmetry, but it should temper it.

Consistent scoring is not the same as a useful test

This distinction gets deliberately blurred, so it is worth pulling apart.

A 2020 study in PLoS ONE ran 105 puppies from 21 litters across 13 breeds through a seven-week behavior test with three independent assessors, and found high agreement between them on four of the five tasks.

Three people watching the same puppy gave the same score. The score still does not predict what the dog becomes. The authors say so in the same paper: the literature “seem[s] to confirm that conducting behavioral tests at 7 weeks of age is too early to reliably predict the temperament and personality of a dog.”

So when a test is described as “standardized” or “consistent” or “validated for reliability,” ask what it is reliable at. Reliability means the graders agree. Validity means the result predicts something. They are different words and only one of them is worth paying for.

So when does assessment start working?

Around five to twelve months, and best in the adolescent window — and even then it is a probability, not a verdict.

The strongest predictive work in this field evaluated dogs at an average age of 58 weeks, roughly thirteen months. That study, in Frontiers in Veterinary Science, is the same one that produced the well-known washout figures — release rates for behavioral reasons of 58 to 60 percent across populations of 3,569, 5,967 and 2,990 dogs.

Its prediction models reached an AUC of about 0.71, with overall accuracy around two-thirds.

That is the best-validated result in the field, on thousands of dogs, using the best instrument available — and it is a long way from a pass or fail. The authors also note the models were built inside a single working dog population and may not generalize elsewhere.

A separate study of 1,401 dogs assessed at five, eight and twelve months, published in PLoS ONE, found trait scores correlating across those ages in the range of 0.42 to 0.69 — usable, and clearly better than anything available at seven weeks.

Something real is measurable at eight weeks. It is just not this.

To be fair to the other side of the argument: a study of 375 eight-week-old puppies published in Current Biology found that puppies were already skilled at following human gestures with no evidence that it required learning, and that “over 40% of the variation in dogs’ point-following abilities and attention to human faces was attributable to genetic factors.”

Social attentiveness to humans is present, heritable and measurable at eight weeks. That is a genuine finding and it is not nothing.

But it is a finding about a specific cognitive skill, not a prediction that this puppy will qualify as a service dog two years from now. Those are different claims and the second one is the one being sold.

What actually predicts success — and what it predicts success at

These are traits with real assistance dog outcome data behind them, not general good-dog traits.

The distinction almost every article gets wrong

Those traits predict whether a dog finishes a training program. That is not the same as whether the partnership works for the handler.

A 2022 study in PLoS ONE of 82 veteran and service dog pairs found that most service dog characteristics did not predict the veterans’ outcomes at all. The one that did was lower excitability toward an approaching stranger.

Program completion is the outcome nearly every study measures, because it is the outcome organizations record. It is a proxy. Keep them apart.

There is no such thing as the right service dog temperament

A 2026 study in Frontiers in Veterinary Science followed 1,760 dogs across six assistance dog organizations in the US and Canada — 678 placed, 1,082 released, an overall success rate of 47.5 percent, which the authors describe as in line with the industry average.

What it found is that different assistance careers select for measurably different temperaments. The profile that suits a medical alert dog is not the profile that suits an autism assistance dog, which is not the profile that suits a mobility dog — and mobility work adds physical requirements on top of temperament that no behavior test measures at all. The direction of those differences is clear even where the exact magnitudes come from small subgroups and should not be quoted as headline figures.

The authors put the consequence well: “Career fit can be a source of stress, particularly when an AD’s placement does not align with their natural temperament.”

So “does this dog have a service dog temperament” is the wrong question. The right one is “does this dog suit this work, for this person.” If you do not yet know what tasks you need, that is the thing to settle first — start at do I qualify for a service dog, and if the dog in question is one you already own, how do I qualify my dog as a service dog.

Breed is not temperament, and the numbers are not close

A 2022 study in Science found that breed explains just 9 percent of behavioral variation between individual dogs.

Nine percent. Which means that when someone tells you a breed has the temperament for this work, they are describing something that accounts for less than a tenth of what is in front of you.

Breed does shift the odds at population level — a 2020 study of Finnish dogs found Labradors “seldomly fearful”, with 0.4 percent showing stranger-directed aggression against 10.6 percent in Miniature Schnauzers. That is a real difference and it is why programs breed Labradors. It is also why the breed lists aimed at autistic handlers are mostly beside the point, and that article works through the same number.

But you are not selecting a population. You are selecting one dog. Which breeds tend to do this work and why is covered in best service dog breeds, and the reason no breed is excluded by law is in can a pit bull be a service dog.

The raiser matters, and that is good news

A study of 978 prospective guide dogs from a single organization — all with matched early environments, all placed with puppy raisers at seven to eight weeks — found measurable effects of the raising environment on aggression and fear at twelve months, published in Frontiers in Veterinary Science.

Temperament is not a fixed quantity you either detect at eight weeks or miss. What happens between eight weeks and twelve months changes the dog. Which is a reason to be skeptical of the early test, and also a reason not to despair over a mediocre one.

Rescue dogs: the honest answer is that nobody knows

People ask this constantly and the internet answers it confidently in both directions. Neither confidence is earned.

We could not find a single peer-reviewed study directly comparing success rates for shelter or rescue-sourced assistance dogs against purpose-bred ones. The only primary work on shelter-sourced service dog candidates is roughly twenty-five years old and behind paywalls.

What can be said honestly: every large predictive dataset in this field is built on purpose-bred program dogs. The models were developed inside those populations, and the researchers themselves caution against assuming they generalize.

So an owner-trainer assessing a rescue dog is working outside the range of every validated model that exists. That is not an argument that rescue dogs cannot do this work. It is an argument that anyone quoting you a success rate for rescue dogs is quoting a number nobody has measured.

There is no official test, and the law does not want one

This surprises people, and it is the single most exploited gap in this subject.

There is no federally required, regulated or legally recognized temperament test for service dogs in the United States. No register of tested dogs. No accreditation with legal force. Nothing to pass.

The Department of Justice, on ada.gov, is explicit that you may train the dog yourself: “People with disabilities have the right to train the dog themselves and are not required to use a professional service dog training program.” And that no paperwork may be demanded: “Covered entities may not require documentation, such as proof that the animal has been certified, trained, or licensed as a service animal, as a condition for entry.”

On the businesses selling exactly that:

“There are individuals and organizations that sell service animal certification or registration documents online. These documents do not convey any rights under the ADA and the Department of Justice does not recognize them as proof that the dog is a service animal.”

What the law regulates instead

Not the dog’s tested temperament. The dog’s behavior, in the moment, every moment.

A service animal may be removed on two grounds only: the dog is out of control and the handler does not take effective action, or the dog is not housebroken. That is a standard you keep meeting, not a gate you pass once. The full picture is in ADA service dog laws, and the industry built on pretending otherwise is dismantled in how to identify a fake service dog.

There are voluntary standards. The International Association of Assistance Dog Partners recommends a minimum of 120 hours of training over six months or more, with at least 30 hours in public, plus behavioral criteria including no aggression, no soliciting food or attention while working, and calm work on leash. IAADP says plainly that failing to meet them does not make a dog illegitimate under the ADA. They bind members of that organization and nobody else.

So what should you actually do?

Test. Just stop treating the result as a verdict, and stop paying for it at seven weeks.

1. Treat a good early result as encouraging and a bad one as almost uninformative. That is what the evidence supports, and it is the opposite of how these tests are sold.
2. Reassess in the adolescent window, somewhere between five and eighteen months. That is where the predictive value actually lives.
3. Reassess more than once. Agreement is strongest between evaluations three to six months apart. One snapshot is one snapshot.
4. Screen hard for the disqualifiers. Aggression is the one trait that is stable early, and it is the one that ends the conversation. Use the test for what it is good at.
5. Watch what the dog does when a task becomes impossible. Looking to a person for help is one of the few things that replicated across independent assistance dog samples.
6. Ask who is running the test and what they are selling. A test administered by the person selling you the puppy is not an assessment. It is part of the sales process.
7. Assess in the places the dog will actually have to work, not in a living room. Thirty of IAADP’s recommended 120 hours are public outings for exactly this reason. What the law says about a dog in training being in those places varies by state and is covered in can service dogs in training go anywhere.
8. Budget for the possibility that it does not work out. Six dogs in ten do not make it, even inside professional programs. If you are training your own, that risk is yours — what training your own dog costs covers what that means in money, and how much a service dog costs covers the alternative, which for most people is free. If the wait for a program dog is what pushed you toward training your own, how veterans get their own service dog sets out how the accredited route actually works and what it costs, which for many people is nothing.

Frequently asked questions

Do puppy temperament tests predict service dog success?

Barely. In one study of 465 puppies, those who passed a standardized test certified as guide dogs 66 percent of the time, against 64 percent for the whole sample including the dogs the test would have rejected. A large meta-analysis found personality consistency in puppies at r = 0.30, meaning the puppy measurement accounts for around nine percent of the variation in the later one.

What age should a service dog candidate be temperament tested?

Adolescence, not puppyhood. The strongest predictive research evaluated dogs at an average of 58 weeks, about thirteen months, and reached an AUC of around 0.71. A study of 1,401 dogs assessed at five, eight and twelve months found usable consistency across those ages. Assessments taken three to six months apart agree most strongly, so reassess rather than relying on one snapshot.

Is there an official temperament test for service dogs?

No. There is no federally required, regulated or legally recognized temperament test, certification or standardized assessment for service dogs in the United States. The Department of Justice states that certification and registration documents sold online do not convey any rights under the ADA. The law regulates the dog’s behavior in the moment, not its tested temperament.

What temperament traits do service dogs need?

The traits with real outcome data behind them are low distractibility, low excitability, low stranger-directed fear, low reactivity to sudden noise, tolerance of physical handling, low prey and chase response, high trainability, and a tendency to look to a person for help when a task becomes impossible. Low excitability toward an approaching stranger is the one trait that predicted outcomes in veteran and service dog pairs.

Can a rescue dog become a service dog?

Legally yes, and the ADA sets no requirement about where a dog comes from. But no peer-reviewed study directly compares success rates for shelter-sourced against purpose-bred assistance dogs, and every large predictive model in this field was built on purpose-bred program dogs. Anyone quoting a success rate for rescue dogs is quoting a figure nobody has measured.

Does breed determine service dog temperament?

Not much. A 2022 study in Science found breed explains just 9 percent of behavioral variation between individual dogs. Breed shifts the odds at population level, which is why programs breed Labradors, but it accounts for less than a tenth of what is in front of you when you assess one particular dog.

Why does a temperament test predict better when it passes a dog than when it fails one?

Because that is what the data show. In an analysis of 15,374 evaluations across 3,969 Labradors, positive predictive value ran from 91 to 99 percent while negative predictive value ran from 0 to 50 percent. Dogs scoring well early tended to keep scoring well; dogs scoring poorly frequently went on to succeed anyway. A failing result is close to uninformative.

The short version

Temperament testing is real, and used properly it is worth doing. It is just sold backwards.

It is marketed as a filter for finding the dogs that will fail, at the age when it is least able to, measuring the traits it is worst at. The evidence says it is trustworthy when it says yes, unreliable when it says no, and most useful somewhere around a year old.

Ask what the test was validated against, at what age, and in how many dogs. If the answer is a leaflet, you are being sold the packaging.