Two ledgers, one borrowed percentage
The famous figure is true and routinely misdescribed. In the FBI's 2015 review, 257 of 268 trial transcripts contained at least one scientifically erroneous statement about microscopic hair comparison. That is not a 96% laboratory false-match rate. A separate casework study found that mitochondrial DNA excluded 9 of 80 hairs that microscopy had associated. Put the denominators back, and two different failures come into view: what an examiner saw, and what an examiner told a jury it meant.
Tap the claim that travelled. Then inspect what it counted.
The label is still folded shut.
Ledger A · 2015 testimony review
...%
95% Wilson interval: ...
This is a prevalence among selected trial cases. It is not a rate per statement, per hair, per examiner, per conviction, or per laboratory comparison.
Ledger B · 2002 casework cross-check
...%
95% Wilson interval: ...
This is discordance within a selected casework series. It is not a universal false-positive rate for microscopic hair comparison.
The numerator in the first ledger is cases with bad courtroom statements. The numerator in the second is hair examinations contradicted by a genetic exclusion. The percentages cannot be merged because the things counted are different.
The further result
It was not the only signal in the record. Houck and Budowle reviewed 170 FBI hair examinations that received both microscopy and mtDNA analysis between 1996 and 2000. Among the 80 that microscopy associated, mtDNA excluded 9. That cross-check concerns the underlying comparisons, not what an examiner later said in court.
Table 2 reports one mtDNA result as inconclusive and one as insufficient in the microscopy-association row. Choose a transparent treatment. Each answer remains a description of this selected series only.
...%
...
...
Why the control stops at two. The paper reports 6 insufficient and 3 inconclusive mtDNA outcomes across all 170 examinations. Only 1 of each lies inside the 80 microscopy associations. Moving all nine into this denominator would silently import results from other microscopy categories.
The pinned source table
The highlighted cell is the 9 in the second ledger. Row and column totals are recomputed in the browser from this transcription of Table 2.
| Microscopy result | mtDNA association | mtDNA inconclusive | mtDNA exclusion | mtDNA insufficient | Row total |
|---|
“Association” in the mtDNA column means the sequences were concordant. It does not identify one person: maternal relatives share mtDNA, and unrelated people can share a haplotype.
What neither ledger can answer
The 2015 release described an ongoing review. The 268 cases were trials with inculpatory FBI examiner testimony found and reviewed under the project’s scope. They were not a random sample of every hair comparison, every examiner, or every jurisdiction. One case could contain several statements and several hairs, but counted once here.
The 170 examinations were hairs submitted to the FBI for both methods from 1996 through 2000. The paper notes that few hairs excluded by microscopy in ordinary casework were sent for mtDNA. Selection happened before the table, so a Wilson interval describes binomial uncertainty conditional on these observations, not the selection process.
An mtDNA exclusion contradicts common source for the compared samples, assuming valid testing and samples. A non-exclusion does not identify one donor. The paper itself says the 9 exclusions should not be construed as a false-positive rate for microscopy or a false-exclusion rate for mtDNA.
The 257 cases contained scientifically erroneous statements. That count does not establish 257 wrongful convictions, legal materiality in each case, innocence, or how often the microscopic associations themselves were incorrect.
A live institutional disagreement
The FBI review page says microscopic comparison remains valid when used with appropriate limits and mtDNA. PCAST’s 2016 review of the studies cited by DOJ concluded that those studies did not provide a scientific basis for calling the process valid and reliable. This page does not settle that broader dispute. It exposes the two selected datasets that both positions must account for.
The instrument has teeth
The page computes its displayed percentages and Wilson intervals from the source counts. The checks below take separate arithmetic routes, compare against source anchors chosen before calculation, and feed both checkers known-wrong data that must be rejected.
Running checks...
Sources and apparatus