Most dogs have the nose for detection work. Few have the temperament, and the formal alert passes along only a small part of what the nose picks up.
What makes a good detection dog? Mostly temperament: a dog that stays confident in new places, around noise and on strange surfaces, and keeps working for a reward. When the Air Force shops for military working dogs, it tests hundreds of candidates and brings home fewer than half, and many of those then fail training. Across working-dog programs, success rates around 50 percent are typical, according to a 2021 review.
Dogs carry about 811 working olfactory receptor genes, roughly twice the human count of 387. A 2025 study of 635 dogs from 121 breeds found individuals carrying between 779 and 807, and scent hounds didn't carry more than other breeds. Another 2025 study found no difference in receptor gene counts between scent breeds, other breeds and short-nosed breeds. Its authors suggested that breed differences in performance more likely come down to motivation and trainability.
Behavior studies point the same way. In a 2015 experiment, pugs learned an odor discrimination faster than German shepherds and held up better as the odor got weaker. Nine of the 10 greyhounds tested didn't stay motivated long enough to finish.
Noses do vary. Detection thresholds varied widely among ten identically trained dogs, scent breeds beat short-nosed breeds on the hardest level of a natural search task, and pugs have lost some receptor genes. Noses matter at the margins, but most documented washouts are behavioral.
Federal law defines the behavioral standards for explosives detection dogs to include "canine temperament, work drive, suitability for training." An Air Force handler told a school group in 2020 that "we need dogs to be confident whenever they are in the field" and that "we don't want a dog that is afraid of heights or loud noises."
Much of the detailed data comes from Auburn University's breeding program. There, 20 percent of 146 purpose-bred dogs washed out, not counting medical releases, and 62.5 percent of those failed on environmental soundness: how a dog copes with new places, surfaces and noise. In a later study of 60 puppies, environmental sensitivity predicted which ones would succeed at every age tested, and work-drive scores didn't. A 2020 review by the same group named poor environmental soundness as the main reason dogs fail.
Temperament is also hard to see in advance. The military's own behavioral tests barely predicted which dogs would certify and didn't improve predictions of detection certification at all, though Auburn's puppy tests did better.
Results vary widely by source. Sweden's dog training center tested 2,107 dogs for police, guide and other service work between 1983 and 1991 and rejected about three in four. Japan's customs service trains 30 to 50 percent of its detection candidates successfully. Even programs that breed and raise their own puppies report results from roughly a quarter to about 80 percent.
Behavior also ends careers. Among 1,230 military dogs discharged in fiscal 2019 to 2021, physical problems ended about half of careers. Fear and anxiety accounted for 9 percent of discharges and the shortest careers, about 3.4 years.
Police and military dogs face a second filter, because most of them also have to bite. One county sheriff's K-9 unit we talked with this year said three of the last nine dogs it trained washed out, mostly in apprehension work rather than detection.
A military dog spends 120 training days at Lackland, 65 of them on detection. After certification the training never stops. GAO reported that the TSA requires its teams to log "240 proficiency training minutes every 4 weeks," and the SWGDOG guidelines recommend at least 16 hours a month per team.
New threats need new training, and what a dog learns doesn't always transfer. Dogs trained on a pure oxidizer lost track of it once it was mixed with other ingredients, as it would be in a homemade explosive. In one study, 87 percent of military dogs trained on pure potassium chlorate missed at least one of four chlorate mixtures. Dogs trained on a fentanyl mimic didn't reliably respond to real fentanyl. Once dogs learn the right thing, they remember it well: one group learned 40 odors over 16 months and still recalled ones they hadn't met in a year.
The work also wears on the sense. Dogs that went six weeks without finding a target in their work area lost search vigilance, which is one reason handlers plant training aids. Exercise makes dogs pant, and panting cuts sniffing. In a small 2024 study, hit rates on the faintest odor fell from 0.87 at a walk to under 0.45 at a trot. Another found that heat and humidity made the detection limit for C4 3.5 times worse.
A detection dog's formal alert is a trained behavior, usually a sit, a down or a stare. It tells the handler that something on the training list is probably here. It doesn't say which substance, how much, or how sure the dog is, and odors that aren't on the list go unreported. Experienced handlers also read changes in the dog's behavior before the alert, but that read lives in the handler's head.
That signal also passes through a person. In a 2011 study, 18 handler and dog teams searched rooms that held no target scent at all. They called 225 alerts anyway, only 21 of 144 runs were clean, and the alerts clustered where handlers had been led to expect a target. A 2020 study of 59 teams, most of them sport teams, found that telling handlers how many hides to expect didn't change false-alert rates, so handler influence varies.
In the field, an alert also can't tell "drugs are here" from "drugs were here." A review of New South Wales police dogs found drugs after 26 percent of 10,211 indications. In 59 percent of the no-find cases the person admitted some contact with drugs, sometimes months or years earlier, and the Ombudsman declined to count those as accurate alerts.
Some of what the dog detects never reaches the alert. Dogs sniffed longer at samples they failed to alert on than at truly blank ones. In a 2025 lab study, still a preprint, signals from the seconds before each search predicted 85 percent of misses, which could warn handlers when a dog is fading. Dogs can be trained to report more, such as telling a stronger concentration from a weaker one, but that takes extra training and the skill can be fragile.
One answer is to produce more of the rare dogs: breed more carefully, select earlier, and raise puppies the way the most successful programs do. The other is to ask less of each dog, by not making every detection dog a bite dog, by training on mixtures that look like real threats, and by reading more of the dog's response directly, from sniffing patterns and heart rate up to the olfactory signal itself. Researchers have decoded odor identity from neural activity in mice and told explosive vapors apart using neural activity in locusts. Doing the same in a working dog is still an open problem.
General Sense exists to remove the bottleneck of having to find the top 1-2% of the general population of dogs to be part of the detection program. We're building a way to read what a detection dog's nose picks up directly from its olfactory system, rather than through a trained behavior. The goal is a dog that needs less training per target, reports which target it found and roughly how much, and widens the pool of dogs that can do the work. Instead of finding the top 1-2% of the general population of dogs, we can source from the top 10-20% of dogs, which will widen the bottleneck by 10-fold.
More importantly, we can automatically log what the dog detects, with GPS and time stamps. This removes a lot of manual paperwork on the handler and the agencies that buy detection services.
What makes a good detection dog? Mostly temperament: confidence in new places, around noise and on strange surfaces, plus the drive to keep working for a reward. Differences in noses between candidates matter less.
Can any dog be a sniffer dog? Most dogs can learn simple odor searches, as pet nose-work classes show. Few have the temperament to work in airports, ports or war zones.
What percentage of detection dogs fail training? About half is typical across working-dog programs. Success rates range from roughly a quarter to about 80 percent, depending on where the dogs come from.
How do detection dogs tell their handler they found something? With a trained final response, usually a sit, a down or a stare at the source.
Do detection dogs make mistakes? Yes. Fatigue, heat, handler cues and leftover odor all produce misses, or alerts with nothing to find.
