(FORTUNE Magazine) – The first thing most people notice about TinkerBelle is her nose. A female bloodhound, TinkerBelle has a fabulous snout, and like any dog, when she isn't sniffing garbage or a chew toy, she might sniff you in places you'd rather not be sniffed. It could be worse. Some of the people TinkerBelle sniffs end up in jail. That's because TinkerBelle is an expert for the South Pasadena Police Department and the FBI in what's called a "dog scent lineup."

Dog scent lineups have played a role in thousands of criminal cases in the U.S. since 1923. In a scent lineup, dogs match the smell on an item from a crime scene to a suspect in a group. Though dogs don't actually perform in court (anymore, at least), their scent identifications are admitted as evidence in most states. People have been convicted of robbery, rape, and even murder when the primary evidence against them is, effectively, a bark.

You would think that the scientific and legal work supporting a dog's ability to actually do this--match odors from all kinds of objects (clothes, doorknobs, even a bullet casing) to a criminal days or months later, in a room filled with other smells--must be pretty strong. It's not. In fact, the most comprehensive scent-lineup studies done so far show that highly trained dogs are wrong a whopping 30% to 40% of the time.

That's dangerous stuff when it finds its way to the courtroom. Just ask Jeffrey Allen Grant, a softball coach whom TinkerBelle identified as the Belmont Shore rapist in 1999. She followed a scent from a crime scene to his door and later picked him from a crowd at the Long Beach Police Department. Grant was arrested and imprisoned--and then, three months later, proved innocent by DNA tests and awarded $1.7 million in damages. (The real Belmont Shore rapist was caught in 2002.) "I've been studying dogs a long time," says I. Lehr Brisbin, a scientist at the University of Georgia, "and when I test dogs that are supposed to be able to do this very well, they fail. Invariably."

What's going on? Aren't dogs made for this kind of work? According to Charles Wysocki, a scientist at Monell Chemical Senses Center in Philadelphia, dogs have something like 250 million olfactory receptor cells, more than most other animals. In the same way that humans make visual images and memories of what they see, dogs get through life making "scent pictures." This quality makes dogs very useful for narcotics work (where they identify the same smell over and over) and in cases where they're trailing a "hot" scent (say, from a fresh crime scene or in search-and-rescue jobs).

But expecting a dog to take a step back and piece together a scent puzzle in a lineup might be one cognitive step too many, according to Brisbin. Because dogs' noses are so sensitive, they are able to differentiate among smells from different parts of the body. But asking them to connect the odor your finger leaves on a bullet casing with the full range of smells you emanate in person is a more complicated association. "It's like showing a child the color red," Brisbin says, "in hopes that he'll return with something orange because it contains red."

Whoa, doggy. Handlers can also confuse their dogs by unintentionally and unconsciously "communicating down the leash," according to Larry Myers, a biologist at Auburn University in Alabama. Tests have shown that in addition to responding to the tiniest tugs of the leash, dogs also pick up on note taking and even on subtle facial movements in hope of pleasing the handler by picking the "right" answer (which the handler shouldn't know but sometimes does: Myers has encountered lineups where, for example, only one guy is handcuffed and wearing a prison jumpsuit).

Even assuming that a dog's nose is reliable and that the animal understands what's being asked of it, there are questions about technical errors in scent lineups. For example, does the controversial Scent Transfer Unit, on which dog scent lineups often stand, actually work? The STU 100, as it is known, operates something like a crime scene dustbuster: It vacuums scent onto a gauze pad, which is then placed in a plastic bag, frozen, and stored in a "scent bank" where it's kept until trial time, when the pad is thawed and offered to a dog as evidence. The STU 100 isn't approved by either of the national bloodhound associations, and scientists like Auburn's Myers have been flying around the country testifying against it, but it is still used by groups including the Southern California Police Bloodhound Handlers Coalition.

Finally, there's the potential for human error and worse. If you know dog owners, you probably know at least one who believes his animal can do no wrong. Handlers--who live and work with their canine partners--aren't much different, and that can lead to less than objective testimony. There have even been cases of possible handler fraud. The most recent example is the case of Sandra Anderson and her dog, Eagle. The pair had an international reputation for turning up evidence in difficult cases. Then, in 2002, Anderson was caught allegedly planting human remains at a crime scene for her dog to "discover." She was arrested on federal charges of obstructing an investigation. Though she denies the charges, the FBI is said to be pursuing up to 50 cases she may have tainted.

None of that would matter if we weren't predisposed to treat a dog scent lineup as something more than the "corroborative" testimony it's supposed to be. But there's just something about a dog's word, isn't there? From Lassie to Rin Tin Tin, all the way back to Odysseus' dog, Argos (the only one to recognize Homer's hero after his 20-year absence), our culture is littered with testaments to the honest dog and his reliable nose. As a result, "people have an inflated idea of what kind of feats dogs are actually capable of," says Russ Hess, president of the U.S. Police Canine Association. There's no doubt that dogs are extremely useful in police work, but relying on them for case-closing testimony can be dangerous. "Juries tend to really like it," says Ted Hamm, a California handler who worked more than 300 calls and testified in dozens of cases last year for the L.A. County sheriff's department. "After a day of legalese, it's a real break for these folks when I start talking about something as low tech as a dog. It's like, 'Oh. I've got one of those at home. I can deal with this.' "