Two humans will have the vast majority of their DNA sequence in common. Genetic fingerprinting exploits highly variable repeating sequences called microsatellites. Two unrelated humans will be likely to have different numbers of microsatellites at a given locus. By using PCR to detect the number of repeats at several loci, it is possible to establish a match that is extremely unlikely to have arisen by coincidence.
Genetic fingerprinting is used in forensic science, to match suspects to samples of blood, hair, saliva or semen. It is also used in such applications as studying populations of wild animals, paternity testing, identifying dead bodies, and establishing the province or composition of foods. It has also been used to generate hypotheses on the pattern of the human diaspora in prehistoric times.
In the early days of the use of genetic fingerprinting as criminal evidence, juries were often swayed by spurious statistical arguments by defence lawyers along these lines: given a match that had a 1 in 5 million probability of occurring by chance, the lawyer would argue that this meant that in a country of say 60 million people there were 12 people who would also match the profile. This was then translated to a 1 in 12 chance of the suspect being the guilty one. This argument is not sound because it implicitly assumes that the suspect is drawn at random from the population of the country. In fact, a jury should consider how likely it is that an individual matching the profile would also have been a suspect in the case for other reasons. The science was made famous in 1994 when prosecutors heavily relied on and through expert witnesses exhaustively presented and explained DNA evidence linking O. J. Simpson to a double murder.
Nowadays, more testing is carried out so that the theoretical risk of a coincidental match is 1 in many billions (??more exact figure??). However, the rate of laboratory error may be much higher than this, and often actual laboratory procedures do not reflect the theory under which the coincidence probabilities were computed. For example, the coincidence probabilities may be calculated based on the probabilities that markers in two samples have bands in precisely the same location, but a laboratory worker may conclude that similar -- but not precisely identical -- band patterns result from identical genetic samples with some imperfection in the agarose gel. However, in this case, the laboratory worker increases the coincidence risk by expanding the criteria for declaring a match. Recent studies have quoted relatively high error rates which may be cause for concern (references??). The cautious juror should not convict on genetic fingerprint evidence alone if other factors raise doubt.
Recently, an additional technique for genetic fingerprinting has been introduced: AFLP, or amplified fragment length polymorphism. This new technique is similar to RFLP analysis, but introduces a few other features, like two rounds of amplification and specially made primers. AFLP analysis is now highly automated, and allows for easy creation of phylogenetic trees based on comparing individual samples of DNA.
See also: DNA testing, Prosecutor's fallacy