Standards and reliability in fingerprint identification

A defendant is appealing his conviction by challenging the fingerprint evidence used against him. His first objection is that the examiner couldn't name studies proving that no two people have identical fingerprints. That's worth separating from the underlying question. Being unable to name a study on demand doesn't establish that no study exists or that the identification is wrong. But criticizing that objection wouldn't, by itself, prove that fingerprints are reliable either. We still need the evidence.

The defendant raises two more concrete concerns: we lack established error rates, and examiners don't use uniform, objective standards. In particular, there is no generally accepted number of matching features required to identify someone. If one agency wants five matching features and another wants ten, are they making comparably reliable judgments? That hypothetical illustrates the standards problem; it doesn't tell us how many features actually matched in this case.

The third paragraph changes speakers. We are now hearing the court's response. Fingerprints have been used for a century, experts have long regarded them as reliable, and the court doesn't think the defendant has supplied a reason to reject this entire kind of evidence. Further testing and more consistent standards could still be desirable. Accepting the evidence is not the same as declaring that nothing could be improved.

The final paragraph explains why the court thinks there is enough consistency already. Agencies may use different thresholds, but examiners follow a common points-and-characteristics approach. Training, proficiency testing and peer review impose standards on them. The court relies on that structure to accept testimony about a very low error rate. Notice the judicial language: “this court” evaluates what the “trial court below” was entitled to do. We're reading the reasoning of a court reviewing an earlier trial.

A's main point: The trial court was entitled to credit fingerprint evidence as highly reliable despite objections about testing and differences in identification standards.

B attacks that confidence. It begins with the standards problem, but gives it a different significance. Some examiners count matching ridge characteristics; others use a holistic approach. Even within point counting, there is no fixed agreement about how many similarities establish a match. B sees no adequate objective rule for deciding when similarities amount to identification.

Then B asks what those similarities actually tell us. Different people can share some features, so how likely is it that another person would share four, five or eight? That matters especially when the crime-scene print is only a smudged fragment. Think of touching the edge of a phone: the mark might contain only part of a thumbprint. A claim about complete fingerprints doesn't automatically tell us how rare that little fragment is.

A one-in-a-thousand chance and a one-in-a-billion chance would give us very different reasons for confidence. B's complaint is that examiners cannot decisively supply the relevant probabilities. We shouldn't fill that gap with our own guess that a mistaken match must be fantastically unlikely.

Finally, B asks how examiners perform in practice. Some tests report few or no errors, but critics say those tests are too easy. A more rigorous test found a 34 percent error rate. That is evidence for concern, not proof that exactly 34 percent of all real criminal identifications are wrong. B uses it to challenge the confidence placed in existing testing.

B's main point: Fingerprint identification faces fundamental problems involving inconsistent standards, unknown probabilities of coincidental matches and insufficiently rigorous error testing.

Both accounts care about reliable identification. Their disagreement is whether existing practices and evidence justify confidence: A accepts the safeguards and low-error testimony, while B argues that serious questions remain.