Differences between traditional linear regression and a Bayesian approach to classification are discussed. Classification consists of assigning subjects to one of several available treatments on the basis of their test scores when the success of each treatment is measured by a different criterion. Formulating this problem as an empirical Bayes decision problem is indicated. As an example, the case of classification with a threshold utility function is analyzed and optimal assignment rules are derived. Data from over 15,500 students taking the achievement test of the Dutch National Institute of Educational Measurement at the end of grade 6, usually used to get an impression of the predictive validity of the test, were used as an example to explore the possibility of establishing optimal rules for the selection of continuation schools using this theory of classification. The following types of secondary education were selected as "treatments" in the analysis: (1) lower vocational education; (2) lower general education; and (3) middle general education. Students who continued at the same school or a higher one were considered successes. Findings were that differences between the two types of general education are very small. Three tables give data for the example study, and one graph illustrates the logistic regression lines for the three treatments. (SLD)