Nine predictor measures available from student files were used in classifying seventh- and eighth-grade students in mathematics classes into one of three or four levels of instruction in the following year. These measures were scores on tests of language intelligence, nonlanguage intelligence, general intelligence, reading vocabulary, reading comprehension, arithmetic concepts, arithmetic applications, and teacher recommendation. At the end of the year, teachers were asked to judge the appropriateness of students placement. The sample consisted of 505 students from a large suburban junior high school in the southeast. A quadratic multivariate classification rule was used for both internal and external analyses. Internal discriminatory power was high; external power dropped for across-groups classification but not appreciably for separate large groups. One predictor measure, based on teacher recommendation, proved to be about as accurate as all nine, in an external sense. (Author/SD)