The best-match problem involves finding the objects in a file that are closest to some specified object. Two major issues are the choice of an appropriate measure of closeness and efficient implementation. This paper addresses the second issue. It presents an algorithm that can be used to avoid an exhaustive search; the algorithm extends and generalizes earlier work. The algorithm is restricted to measures that are distance metrics, so it cannot handle some similarity and dissimilarity measures. For most applications, this restriction is reasonable. The algorithm assumes that distances are precomputed for an arbitrary set of pairs of objects. This limited distance information is used to determine upper and lower bounds for other distances. Many items can be eliminated from further consideration, and exact distances can be computed only as needed. The authors present experimental results for both protein data and randomly generated data for files containing 100–300 items. The paper is well written and should interest those involved in research or system implementation in this area.