By Mark Derr, NYTimes

MIAMI (NYT 23DEC02) - When bomb-sniffing dogs indicated the
presence of explosives last summer in the cars of three medical students
bound for Miami, the authorities detained the men and closed a major
thoroughfare across South Florida. No trace of explosives was found in their
cars.

Now, a number of scientists and trainers are expressing concern that such
mistakes could become more common as thousands of new canine detectives are
deployed across the country.

Experts on explosives detection say that when dogs' handlers are excited and
stressed, the dogs may overreact and falsely suggest that explosives are
present when they are not. False alerts are better than missing a live bomb,
they say, but it is better for the dogs to be accurate.

More rigorous training and certification standards and more research into
the way dogs detect scents and the relationship between them and their
handlers are needed to avoid these problems, said Dr. Lawrence J. Myers, an
expert on dog olfaction at the Auburn University College of Veterinary
Medicine.

Dogs are far better at sniffing out the source of a particular odor than any
machine yet developed, experts say. They are also more manageable and
culturally acceptable than rats and other animals adept at detecting scents.

Scientists have estimated that a dog's nose has about 220 million
mucus-coated olfactory receptors, roughly 40 times as many as humans.

When a dog sniffs, chemical vapors - and, perhaps, tiny particles - lodge in
the mucus and dissolve, sending electrical signals along the olfactory nerve
and ultimately to nearly all parts of the brain. In dogs, the vomeronasal
organ in the roof of the mouth and two branches of the trigeminal nerve in
the nasal cavity also play roles in scent detection.

Skilled trainers have taught dogs to detect just about anything that emits
even the faintest odor, including explosives, underground oil and water
leaks, contraband food, termites, guns, drugs and cash. But in most cases,
scientists have not measured the lowest levels of odor that dogs can detect.

Training and handling dogs is an art at which some people excel, and
together top dogs and top handlers can perform extraordinary feats. But
there are limits on dogs' performance that are frequently overlooked. Poor
handlers alone, Dr. Myers said, can cause dogs' vaunted accuracy rate of 85
percent to 95 percent to plummet to 60 percent, Dr. Myers said.

"Dogs want rewards," he added, "and so they will give false alerts to get
them. Dogs lie. We know they do."

Determining how accurately dogs in general detect particular odors is
difficult, experts say, because procedures vary from place to place, and few
have been subjected to rigorous scientific testing. Though some dogs and
handlers are consistently good, all may vary in their daily performance.

When dogs are asked to identify people, the situation is even more complex.
This use of dogs is based on assumptions that every person has a unique
scent, that odor is stable over time and that dogs can tell one person from
another. But the first two assumptions have not been fully verified and the
last is not always true, said Dr. Adee Schoon, scientific adviser to the
Netherlands National Police Agency Canine Department.

"You need special handlers and special dogs for identifying suspects," said
Dr. Schoon, who recently visited Florida International University in Miami
to present a seminar on scent identification at its International Forensic
Research Institute and to discuss collaborative research.

In the 1990's, Dr. Schoon documented that the Dutch procedure for
identifying suspects with dogs was prone to substantial errors. Then, she
redesigned it.

Her biggest achievement, she said, has been discrediting people who say, `My
dog never errs.' "

Scent identification in Holland is now conducted under controlled
circumstances to minimize human and dog errors. Investigators ask the
suspect and six "foils," who have had no involvement in the crime, to hold
small steel tubes briefly.

The tubes are then lined up on a platform in parallel rows of seven each in
a pattern unknown to the handler. The dog's task is then to match a scent
from the crime scene to tubes in two rows.

The dog performs two tests, the first to prove that its nose is on target
and that it has no interest in the scent of the suspect, by tracking down
the tubes touched by a foil.

In the second, it identifies the suspect, if that person's scent is present,
from scents taken at a crime scene. The dog works off its leash to minimize
the handler's influence. "All kinds of problems" arise when a dog is asked
to match scent to an actual person, Dr. Schoon said. For one thing, she
said, the handler may unconsciously direct the dog toward a particular
suspect.

Dogs are also known to become fixated on people for no apparent reason and
to return to them again and again, Dr. Schoon said. Without the first test
run in which the dog is asked to find another "suspect" in the same group,
it is very difficult to tell when a dog is becoming fixated for no apparent
reason, she added.

Dr. Steven J. Hatfill, a bioweapons expert, was identified by federal law
enforcement authorities this summer as "a person of interest" in the anthrax
inquiry based in part on scent-matching by three bloodhounds.

Pat Clawson, a spokesman for Dr. Hatfill, questioned the circumstances
surrounding the use of bloodhounds and said the dogs' responses were
inadequate to link Dr. Hatfill to the anthrax letters. No charges have been
brought in the case.

Experts say other problems can emerge when the dog is faced with only one
person. Because dogs are regularly rewarded for choosing suspects in
training, they are predisposed to say yes when asked to match scents in
situations involving only one potential suspect, experts say. It takes time
to train an animal to say no in such cases.

Edward Hamm, a member of the Southern California Bloodhound Handlers
Coalition and one of the bloodhound handlers involved in the Federal Bureau
of Investigation's anthrax work, said that all three dogs in the Hatfill
case were specifically trained to say no, but he would not discuss any
details of the case. Paul Bresson, an F.B.I spokesman, also declined to
comment.

Determining the accuracy of detection and scent identification dogs is often
difficult. Certification standards for dogs and handlers vary markedly from
state to state and agency to agency.

Written training logs, which are used to establish a dog's reliability in
court, are themselves often unreliable.

"There is a saying in Holland that the training log is a lie," Dr. Schoon
said, if only because handlers want their dogs to look good. It is not known
how often this problem crops up in the United States.

Dr. Myers said: "The standard measure of a dog's accuracy is what it finds.
The best programs subtract from that score the number of false alerts, but
most do not and so they have no accurate measure of their dogs'
reliability." He is helping to create software to assist handlers and
trainers in selecting, training and maintaining their dogs at optimal
levels.

Maine Specialty Dogs in Alfred trains dogs for fire departments around the
country to search burned out buildings, often for minute traces of flammable
compounds that may have been used in arson, said the head trainer there,
Paul Gallagher. Sponsored by State Farm Insurance, the school selects,
trains, certifies and recertifies about 100 "arson dogs" a year. No dog that
has even one false alert in its final proficiency test receives
certification, he added.

Around the country, a few other programs are equally demanding. Secret
Service bomb dogs, considered among the best in the world, are retested
weekly and must have an accuracy percentage in the upper 90's, said a
spokesman, Brian Marr.

While concerned about missed targets, many trainers and handlers deny that
their dogs sound false alarms, and so they do not record them, especially if
they occur in the field. They argue instead that the dog is picking up a
faint trace of a substance that was once present, or that a handler caused
the dog to err.

Handlers can create errors by pulling their dogs away from things they are
investigating, by letting them search too long in a single place or by
inciting the dog through some gesture, glance or emotion, even unconscious.
Trainers say the message "travels right down the leash."

Mainly for that reason, the few studies of dog performance that have been
done suggest that dogs perform best off their leashes.

Off-leash work is common in Europe, but for a variety of social and legal
reasons, dogs are worked almost exclusively on-leash in the United States,
said Dr. Paul Waggoner, interim director of the Canine and Detection
Research Institute at Auburn.

Other factors can also hurt a dog's performance, Dr. Myers said. He
estimates that in any year, 35 percent of detection dogs temporarily lose
their sense of smell because of illness, tooth decay or other physical
problems.

Weather also affects performance. Dry, hot weather can cause the mucus in
the dog's nose to dry out. Hot, humid weather brings early fatigue. Extreme
cold kills scents, and the wind scatters them.

Creatures of habit, dogs also can become stuck in their ways. For example, a
dog might become fixated on a particular object or smell, Dr. Myers said,
citing a police dog in Alabama that began alerting its handlers to Ziploc
bags because the police stored drug training samples in them.

In the 1990's, researchers at Tel Aviv University showed that dogs would
begin to slack off if they were given fewer samples to sniff for than they
had been trained to find.

The researchers also found that after several days of patrolling an area,
like a stretch of road, the dogs would give up if they discovered no
explosives.

As a result, bomb-sniffing dogs in Israel are continually rotated to new
areas on patrol, said Dr. Joseph Terkel, the professor of zoology at Tel
Aviv University who directed the research, conducted by a doctoral student,
Irit Gazit. Trainers also vary the number, size and concentration of
targets, down to zero, in practice and include "blanks" with different
scents, which the dog should ignore.

Because Israeli bomb dogs work off-leash, 50 to 100 yards ahead of their
handlers and often out of sight, the Tel Aviv researchers several years ago
developed a miniature microphone that fits on the dog's nose and allows the
handler to hear whether the dog is panting or sniffing.

A panting dog cannot sniff. A radio receiver allows the handler to recall
the dog and send it out to search a specific site again if necessary.

The practice of training dogs on substances concocted to replicate the
primary odors found in drugs or explosives can also lead to error, said Dr.
Kenneth G. Furton, director of the forensic science program at Florida
International University.

Studies have shown that different dogs respond to different components of an
odor and that those components change over time. So dogs accustomed to a
concoction used in training may have a hard time recognizing the more
complex bouquet of the actual substance.

Experts say more research may resolve uncertainties and maximize dogs'
performance. Meanwhile, they say, training and certification standards
should be tightened to ensure that dogs and handlers are as reliable as
possible.