Collecting ballistic "fingerprints" from millions of new guns would create a database that would be too unreliable to be useful in solving gun crimes, a team of scientists said Wednesday.

The 300-page report from the National Research Council advised against a proposal that some lawmakers called for in the wake of the 2002 Washington-area sniper shootings.

The concept relies on the assumption that individual guns leave unique markings, like fingerprints, on bullets and shell casings. Some in Congress have said that every new gun should be test-fired so those markings can be entered into a database. Investigators could then use the database to identify which gun fired bullets found at crime scenes.

But the study said the idea is not feasible because digital imaging technology isn't reliable enough to distinguish tiny differences in the markings.

The proposed database would be similar to the National Integrated Ballistic Information Network, which is already in use. Run by the Bureau of Alcohol, Tobacco, Firearms and Explosives, the database includes ballistic data on about 100,000 guns used in crimes. It has about a 75 percent to 95 percent success rate, the scientists found.

With that kind of success rate, supporters say, creating a database for all guns only makes sense.

"Ballistics testing is only as useful as the number of images in the database," Sen. Herb Kohl, D-Wis., said while pushing for the database in 2002.

Actually, the opposite is true, Wednesday's report said. The larger the database, the more errors the computer will return.

Under the current system, the computer might find 10 possible matches for a single bullet and there's a good chance one of them will be confirmed. After adding more than 1 million guns to the database each year, the same system might produce hundreds of possible matches.

"It's a scale problem," said John Rolph, chairman of the group that completed the study. "If we're talking about using this in criminal investigations, we've got to be able to get something that's practically useful."

Neither Kohl nor other lawmakers who supported the database proposal had any immediate comment on Wednesday's report.

The report also questioned the underlying theory behind ballistic imaging. It said the idea that each firearm produces unique marks has not been scientifically proven.

Other variables make the program even more unreliable, the report found. For instance, guns leave different markings on different types of ammunition and the type of ammo used in a crime might not be the same type used during test-firing.

A national database also would not account for the millions of guns already in use. ATF spokesman Robert Browning said the average gun used in a crime is 10 years old. He said the idea of a national database for all handguns needs to be studied more.

"I think what we have now is working well," he said.

The Justice Department contracted with the National Research Council, an arm of the National Academy of Sciences, to conduct the study. Its authors are professors at the nation's top universities and researchers at technology companies.