Controversial study supports admissibility of handwriting.

The notion that each person's handwriting is unique is at least as old as Aristotle, who said, "Just as all men do not have the same speech ... so do all men not have the same writing."

But such views have gotten short shrift in the modern courtroom, where judges have generally disallowed handwriting analysis to establish the identity of a writer.

Now, the classical idea of uniqueness is trying to stage a comeback, thanks to an unlikely ally: the computer.

In the July Journal of Forensic Sciences, computer scientist Sargur Srihari of the State University of New York at Buffalo throws the latest volley in a rancourous debate that is unlikely to end anytime soon.

Srihari's study--hotly contested by defense lawyers and many academics--asserts that with a reasonable writing sample, computers can accurately distinguish individual handwriting with 95 percent accuracy--100 percent with further work by a human analyst. Such claims could bolster not only criminal prosecutions but also civil lawsuits. Plaintiff lawyers rely on handwriting analysis in litigation areas ranging from contracts to medical malpractice.

"We wanted to say in some objective manner that handwriting is individualistic," said Srihari.

That's a tall order for someone with no background in either forensics or law. Srihari first came to national attention 20 years ago when the U.S. Postal Service learned of his doctoral work in pattern recognition. A software program that he created helped streamline mail sorting and can read 75 percent of all handwritten addresses. (The remainder must be sorted by hand.)

But reading a person's handwriting is a fundamentally different proposition from determining who might have written a certain document. Still, the National Institute of Justice--stymied by a string of U.S. Supreme Court decisions that severely restricted the admissibility of handwriting evidence--was impressed enough by Srihari's success at the Postal Service to fund his most recent study.

Handwriting analysis wowed the skeptics in 1935 when document examiners were able to link Bruno Hauptmann to ransom notes in the Lindbergh baby kidnaping. But its influence has waned considerably. The essential problem is that in several rulings, most notably Daubert v. Merrell Dow Pharmeceuticals, Inc. (529 U.S. 579 (1993)), the Court established that "the gold standard is no longer expert opinion, but science," said J. Douglas Peters, a Detroit lawyer who represents plaintiffs in medical malpractice cases. The result was that courts routinely threw out handwriting analysis for lack of a scientific foundation. And "junk science," in the hands of a competent defense attorney, can mean "Whatever is being used to hurt" the defendant, Peters said.

At Buffalo's Center of Excellence in Document Analysis and Recognition, which Srihari directs, researchers developed a four-paragraph document with all letters, numbers, and punctuation that appear in English. Roughly 1,500 study participants--varied in ethnicity, education, age, and sex--copied it three times using a ballpoint pen.

The writing was scanned by computer to find distinctions like slant, spacing, and the formation of "loops." The computer found, with 98 percent accuracy, whether two pages were written by the same person (95 percent when extrapolated to the entire U.S. population). With a smaller amount of information--as little as one word--the accuracy rate dropped to 84 percent.

Even though the study is relatively new, several courts--most notably a federal court in Pennsylvania--have accepted its results. (United States v. Gricco, No. 746037 (E.D. Pa. Apr. 26, 2002).) The academic community, however, has not.

"His claims to uniqueness are truly strange," said Michael Risinger, a professor at Seton Hall University School of Law. "It's a bizarre 19th-century notion." He noted that even DNA researchers don't make claims to uniqueness, relying instead on the probability of a match.

And there are other criticisms: The sample size is too small, the participants too different to draw meaningful comparisons. The source document is too large--often, police or plaintiffs have nothing more than an address on an envelope. Perhaps most important, the study did not address forgery.

Though Srihari acknowledges that his research is in its infancy and will be improved upon in the future, he defends his study. "Usually a data set of 1,000 is considered representative of the general population," he said. "It was a reasonable statistical sample."

For plaintiff lawyers, there may be some documentary problems that a computer may never be able to solve.

"In my experience, we almost never get the documents that are most damaging to the defendant," Peters said. "Those get shredded or burned."