The recent discovery of hypervariable regions of the human genome provides scientists with an important new tool for forensic inference. The DNA data obtained from these hypervariable regions have been dubbedDNA fingerprints. Despite the potential power of DNA fingerprints, their use has been fraught with controversy, deriving in part from a lack of statistical methods to summarize the information contained in a DNA fingerprint. This article provides such a method, useful for forensic inference in criminal (e.g., murder or rape) and civil cases (e.g., paternity). The major difficulty with DNA fingerprint data is that the alleles (distinct DNA fragment sizes) are not observed; rather, one observes measurements of the DNA fragments with substantial error. When measurement error is superimposed on the discrete allele distribution, the fragments follow a mixture distribution. Consequently, our first step for forensic inference was to estimate the mixing distribution. With an estimate of the allele distribution, we summarize the information contained in a particular DNA fingerprint as the likelihood that the suspect and evidentiary samples are from the same individual versus the likelihood that they are from two different individuals. The informativeness of DNA fingerprint loci can be determined using Bayes risk methods. The methods we suggest are illustrated with a data set of DNA fragments from two hypervariable regions of the human genome (two loci, D2S44 and D17S79). Our results indicate that DNA fingerprinting can indeed by a very powerful tool for forensic inference.