In this article
A dog checks three stands in turn, off its lead, sniffing the small glass jar on each. Inside every jar is a strip cut from a worn sock. Behind a one-way screen, the handler has no idea which socks, if any, came from someone with Covid-19, so the dog cannot pick up a hint1.
That was a UK study whose authors call it the first randomised, double-blind trial of its kind, and the dogs did well1. So can dogs really smell disease?
The short answer is yes, sometimes, in carefully controlled tests. The longer answer depends on what happens when the samples are new, nobody in the room knows the answers and the illness is rare.
From anecdote to experiment
The research began with case reports. The first, in 1989, described a dog that seemed to have detected its owner's melanoma, and similar reports followed7. Turning stories into evidence means testing trained dogs on samples they have never met.
Most of that research is small. A 2022 systematic review found 62 studies, published between 2000 and 2021, in which dogs or rats were trained to detect cancers and infections in people. Dogs featured in 57 of them, the average study used about four dogs, and every study trained with rewards of food or a toy7.
Two trials where the dogs did well
In the Covid-19 trial, six of seven dogs were judged ready after six weeks of training. The test set was socks from 200 infected people with mild or no symptoms and 200 uninfected people, none of them used in training1. Any mix of samples could sit on the stands, so picking the odd one out would not work1.
A 2025 UK study of Parkinson's disease worked in a similar way. Two young dogs, a golden retriever and a Labrador cross, sniffed skin swabs from 40 people with Parkinson's and 60 without, all new to them, in lines of four holding one positive or none, while neither handler nor experimenter knew which2.
Two encouraging trials
Even here there are limits. Those Covid figures allow for errors in the PCR test itself; judged strictly against PCR, the dogs found 78.5% to 89.1% of infected samples1. Only two of the ten dogs first screened for the Parkinson's study completed training, and every Parkinson's swab came from someone already diagnosed2.
The Parkinson's team write that they "do not envisage dogs being diagnostic"2. With refinements, they think, dogs could help to check new tests and speed up screening, and they call for studies of whether dogs can alert before a diagnosis2.
When the dogs got it wrong
The cautionary tale is a 2014 Cambridge trial on prostate cancer. Of ten dogs recruited from training classes, four were too excitable and three not motivated enough, leaving three to learn to tell cancer samples from controls3. Two seemed to manage it in training, using samples they may have met before3.
Then came double-blind tests on new samples, with one cancer sample in each line of four. Guessing would have found a quarter of them. The dogs found 13% and 25%, no better than chance3. Beforehand, the team and trainers had been convinced that one of them, a nine-year-old yellow Labrador, could detect a general prostate cancer odour3.
The likeliest explanation is that the dogs memorised the smell of each donor's urine in training: the Labrador need only have remembered 12 cancer samples to explain its training scores3.
Dogs may memorise the individual odours of large numbers of training samples rather than generalise on a common odour3.
That does not rule out training dogs to detect prostate cancer, the authors add3. Other teams have hit similar walls. In one lung cancer study, dogs at first learned where samples had been collected, hospital or home, instead of who was ill7.
Why a lab result is not a screening test
Two numbers describe a sniffer dog: sensitivity, the share of real cases it finds, and specificity, the share of clear samples it leaves alone2. Neither answers the question you would ask if a dog sat down beside you: how likely is the alert to be right?
That depends on how common the illness is. In the Covid-19 trial, the researchers calculated that the chance an alert was right would range from 0.4% to 55.9%, depending on how widespread infection was1. Picture 1,000 travellers, 10 of them infected, and a dog that finds nine infections in ten and clears 89% of everyone else, the figures the researchers used in their modelling1. By our sums, it would flag about 118 people, and only 9 would have the virus.
The team proposed dogs as a quick first sift, with a PCR test to confirm every alert, and described their accuracy as measured "under laboratory conditions"1.
Test design matters too. In 38 of the 62 studies in the review, every line held a set number of positives, so an animal could pick the odd one out, and a handler expecting a find could sway the dog without meaning to7. Real screening offers no such guarantee: a line might hold no positive at all7.
An Austrian team tried that harder test in 2016, with breath from 122 smokers and ex-smokers, 29 of them with lung cancer, and zero to five positives in each line8. Only 30.9% of the dogs' alerts were right. Dogs might not be as powerful in real screening as hoped, the authors concluded, and they suggested one main reason: the stress, for dogs and handlers, of so few finds8.
Only six of the 62 studies in the review used a double-blind, screening-like design, and in those, results usually fell at first when testing went blind7. The reviewers rated the evidence as low quality and concluded that medical detection by animals' sense of smell "lacks evidence and performances so far to be applied in the clinic"7.
Hypo alert dogs: real help, with limits
Hypo alert dogs live with people who have type 1 diabetes. They are trained to alert when their owner's blood sugar falls too low, and many start flagging high readings too, so the owner can test and act4.
The UK study, of dogs trained by the charity Medical Detection Dogs, covered about 4,000 episodes of blood sugar running too low or too high, and the dogs varied: ten were below the charity's 75% accreditation standard when the data were collected4. The owners kept the records themselves, so some less favourable ones may have been left out, the authors acknowledge4.
The American study counted alerts during high blood sugar as wrong, which both teams say may have inflated its false alarms4, 5. Its dogs did alert 3.2 times as often during a low as during normal readings, but when a dog and the glucose monitor both caught a low, the monitor would have warned first 73% of the time, by a median of 22 minutes5. Owners still rated their satisfaction at 8.9 out of 105.
How to read the next sniffer dog headline
These questions sort a promising trial from a test you could rely on.
Questions to ask of a new study
- How many dogs were tested, and how many people gave samples?
- Was everyone in the room blind to which samples were which?7
- Were the test samples new to the dogs?3
- Could a line of samples hold no positive at all?7
- Did the samples come from people already diagnosed, or from people who did not yet know?2
- Did the dogs screen real people, or jars in a laboratory?1
- Has a team with no stake in the result repeated it?
If your dog seems to be telling you something
Perhaps your dog keeps sniffing at a mole, or will not leave your side. It is natural to wonder. But even trained dogs in careful trials miss cases and raise false alarms1, 2, and stories of pets sensing illness cannot show that animals are reliable at finding disease, the reviewers point out7.
What matters is the change you have noticed. The NHS advice is to see a GP about any symptom that is not normal for you, such as a new lump or a new or changing mole, and to go for cancer screening when you are invited6. Most such symptoms can have other causes, the NHS adds, but finding cancer early may make it easier to treat6.
If you have type 1 diabetes and like the idea of an alert dog, talk to your diabetes team. A dog's alert is a prompt to check your blood sugar, and on its own it is unlikely to tell a low from a high4, 5. And if it is your dog's own behaviour that has changed suddenly, that is worth a word with your vet.
The idea of a dog that can smell illness is a lovely one, and in a few careful laboratories it is starting to look like more than an idea. At home, let your dog's nose get on with reading the lampposts, and leave the health checks to your GP.
Sources and further reading8 sources
The small numbers in the article link to the source behind each fact. Sources open on their own websites.
- Using trained dogs and organic semi-conducting sensors to identify asymptomatic and mild SARS-CoV-2 infections: an observational study
- Trained dogs can detect the odor of Parkinson's disease
- Key considerations for the experimental training and evaluation of cancer odour detection dogs: lessons learnt from a double-blind…
- How effective are trained dogs at alerting their owners to changes in blood glycaemic levels?: Variations in performance of glycaemia…
- Reliability of Trained Dogs to Alert to Hypoglycemia in Patients With Type 1 Diabetes
- Cancer
- Remote Medical Scent Detection of Cancer and Infectious Diseases With Dogs and Rats: A Systematic Review
- Canine scent detection for the diagnosis of lung cancer in a screening-like situation