To fix human failings, biometrics vendors—who see security in the use of behavioral and physiological characteristics for identification— propose more machines. Buoyed by the events of September 11, the stock of one such company, Visionics, quadrupled by December, suggesting that the markets concur.

With the government boosting biometrics for homeland security, all that remains is to prove that the systems really work. And there's considerable doubt: Privacy and civil liberties groups like the ACLU have lambasted facial recognition technology as inaccurate.

The main problem is a shortage of unbiased performance data. John D. Woodward Jr., a former CIA operations officer and now a senior policy analyst at policy think tank Rand, notes, "One of the challenges for the biometrics industry is that there really is a lack of independent testing and evaluation of biometrics."

For businesses that are eager to jump on the biometrics bandwagon, Beth Givens, director of Privacy Rights Clearinghouse, cautions that errors may quickly lead to litigation. "If a lot of these systems are put in place, I think we'll see a lot of lawsuits from individuals who have been falsely detained," she says.

But at California's Fresno Yosemite International Airport, where a facial recognition security system has been in place since October 26, only two passengers have been detained to date as the result of a false match, claims spokesperson Patti Miller. That's out of the roughly 2,000 she estimates pass through daily.

According to Ron Cadle, a vice president at Pelco, the company monitoring the Fresno airport test run, about 1 percent of passengers scanned generate a false match. However, true matches (picking criminals out of a crowd) appear to be a crapshoot. A recent U.S. Army Research Lab test revealed that the Visionics FaceIt system (which the Fresno airport uses) correctly identified individuals only 51 percent of the time.

Also, the possibility of false matches raises certain questions. "What happens if you're matched and you're innocent?" asks Givens. "What sort of grievance process do you go through? Are you jailed indefinitely until a more thorough background check is done?"

Biohazards
Fingerprints and irises may be unique, but that doesn't mean machines can make sense of them.

Biometric | The Problem
Face | Affected by lighting, age, posture, movement, glasses, and more.
Fingerprint | 2 percent of people lack readable fingerprints—or fingers. Skin condition is also an issue.
Hand | Disney World uses it, but hand geometry is affected by rings, bandages, and enormous Mickey Mouse gloves.
Iris | No false matches in 2 million, but it can't read about 0.5 percent of eyes. Expensive and creepy.
Vein | About 10 percent "equal error" rate, same as facial recognition.
Voice | Affected by illness and age, and much easier to spoof than most other biometrics.