DNA identification can be quite effective if used intelligently. Portions of the DNA sequence that vary the most among humans must be used; also, portions must be large enough to overcome the fact that human mating is not absolutely random. Consider the scenario of a crime scene investigation . . .
Assume that type O blood is found at the crime scene. Type O occurs in about 45% of Americans. If investigators type only for ABO, then finding that the "suspect" in a crime is type O really doesn't reveal very much.
If, in addition to being type O, the suspect is a blond, and blond hair is found at the crime scene, then you now have two bits of evidence to suggest who really did it. However, there are a lot of Type O blonds out there.
If you find that the crime scene has footprints from a pair of Nike Air Jordans (with a distinctive tread design) and the suspect, in addition to being type O and blond, is also wearing Air Jordans with the same tread design, then you are much closer to linking the suspect with the crime scene.
In this way, by accumulating bits of linking evidence in a chain, where each bit by itself isn't very strong but the set of all of them together is very strong, you can argue that your suspect really is the right person.
With DNA, the same kind of thinking is used; you can look for matches (based on sequence or on numbers of small repeating units of DNA sequence) at a number of different locations on the person's genome; one or two (even three) aren't enough to be confident that the suspect is the right one, but four (sometimes five) are used and a match at all five is rare enough that you (or a prosecutor or a jury) can be very confident ("beyond a reasonable doubt") that the right person is accused. (Human Genome Project)
