Variable number of tandem repeat (VNTR) sequences are used to link defendants with crimes by matching DNA patterns. The probative value of a match is often calculated by multiplying together the estimated frequencies with which each particular VNTR pattern occurs in a reference database. However, this method is liable to potentially serious errors because ethnic subgroups within major racial categories exhibit genetic differences that are maintained by endogamy. The multiplication procedure currently in use can be made scientifically valid only by extensive sampling of VNTR frequency distributions in a variety of ethnic groups, similar to the ethnic studies of various blood groups done in the past. Alternative approaches for dealing with subpopulation heterogeneity are discussed.