At worst this is a 1% error rate, at best 0.1%. Scientific validity aside, I find it unbelievable this was not considered reasonable doubt.
If those 1000 pairs of hairs really were from different, randomly choosen people. But we don't know the quality of that sample. Maybe the examiner looked at 1000 pairs of hairs, 10 looked similar to him and those 10 were from different heads, while the 990 that looked different were from the same head. That would give an error rate of 100%.
Even if we assume there were no errors in the sample, then we still don't know anything about the correlation between the judgement of the examiner and reality. Maybe he randomly considers one in 100 pairs of hairs to look similar. So while yes, the error rate would be 1%, his judgement would have absolutely no informative value. It would be as good as throwing dice.
Whilst DNA evidence has subsequently exonerated him, there were pretty good reasons for the jury at the time to consider the case proven beyond reasonable doubt that didn't involve placing too much faith in dubious "expert witness" testimony about hairs.
There are good grounds to be somewhat suspicious of forensic evidence; however, my understanding is that technical evidence like this is usually far, far more reliable than e.g. eyewitness statements. Those are horribly unreliable, and experienced investigators (or should we say "investigators") are able to convince their witnesses of having seen things that did not really happen at all.
Though since they've had DNA techniques sufficient to overturn the expert witness evidence for rather a long time now it's surprising it took this long to overturn.
It's possible the jury wouldn't have convicted him on eye witness testimony alone, but the fact that the physical "scientific" evidence agreed made it seem more legitimate.
This is how I understand it now. It's not that he says that he compares one hair to those 1000, and then still was sure to be able to keep them apart. That was how I first read it.
For a good comparison, he should have used a line-up. Take one hair from 100 persons with similar hair color and use that as test sample. If he was still able to tell which was from the same person, he would have a case.
The problem is courts often misapply statistics. Prosecutors will often say in a case like this that there is a 100:1 chance that he did it because of the error rate.
But they aren't applying Bayes Theorem.