In response to the increasing importance of student performance in required classes, research was conducted to compare two prediction procedures, linear modeling using multiple regression and nonlinear modeling using AID3. Performance in the first college math course (College Mathematics, Calculus, or Business Calculus Matrices) was the dependent measure. Independent measures included demographics, high school achievement data, and College Board test scores. Using "type of mathematics" as either an independent dummy variable or as a moderator variable was considered. The data came from over 3,500 students who were beginning freshmen in 1986 and 1987 and who took one of the three math courses. The models were developed on the 1986 beginning freshmen and were compared to the sample on which they were developed and cross-validated on the 1987 freshmen. The following conclusions were reached: (1) the statistical models were useful in predicting performance, particularly among those who are predicted to have difficulty; (2) actual ability to explain performance "shrinks" from the modeling sample to the use sample; and (3) the 70% unexplained variance in performance calls from the developing of more discriminating critical variables. Data are presented in several tables and three appendices. Contains 17 references. (Author/KM)