Four hypervariable loci were examined in DNA samples of American Blacks and Caucasians. Genetic diversity, measured by mean kinship within race, is 0.004 for a sliding window equal to twice the radius of coalescence of the autoradiographic bands. Kinship increases with the width of a window or bin, but it is an order of magnitude less than for blood groups and isozymes. This must reflect the greater mutability of hypervariable loci and absence of divergent selection. Low kinship implies that matching probabilities do not depend critically on race. Complete matching between pairs of loci shows deviations from independence, presumably because of inadvertent inclusion of replicated samples or related individuals. Multilocus matching probabilities were corrected for this (possibly spurious) dependence; the correction is negligible. When the complexities of coalescence, kinship, and dependent matching are allowed for, the probability calculations used in forensic identification are reliable. Recent counterarguments without theoretical basis or empirical support are discussed.