The nonparametric empirical discriminant function was applied to stream-sediment geochemical data from the Derbyshire and Denbighshire areas in the United Kingdom in an attempt to predict both the stratigraphic affinity and occurrence of mineralization in the local bed rock. The discriminant functions were defined on the basis of small “training sets” of samples collected from the important geologic features of the two areas; we were successful in correctly assigning 63 and 56 percent of the stream sediments in both areas to their proper stratigraphic or mineralization class for the two areas. Log transformation of the geochemical data materially improved the classification rate. The technique seems clearly to be of use in efficiently recognizing (1) bedrock areas of geochemical similarity, despite the presence of mixed lithologies in the bedrock formations and the presence of extensive glacial overburden and (2) localized areas of unusual geochemistry due to manganese enhancement (scavenging?) in the secondary environment or to past mining and smelting activities.