A unified framework for classification and prediction problems is presented. Well known and lesser known relationships among correlations, distances and error rates are established. A new population distance, the shrunken generalized distance, and a new estimator of the actual error rate are introduced. (31pp.)