The process of selecting applicants for admission to a new school of medicine (Southern Illinois University School of Medicine) is examined through the use of multivariate regression techniques. More than 250 variables were analyzed and classified as to source of data: student American Medical College Application Service (AMCAS) application, supplemental application, letters of recommendation, three interviews by members of the faculty and staff. The predictive ability of the classes of variables are examined by each source and again by combination of sources to simulate the order in which data are received by the office of admissions. A curvilinear regression model is utilized to examine the presence of any incremental prediction of the admissions decision that occurs with the addition of squared and cross-product terms. In all analyses, stepwise methods were used to force linear components to enter the regression equation before allowing curvilinear components to enter. When analyzing data from specific sources in isolation, the curvilinear model increases prediction as measured by the squared multiple correlation coefficient in approximately half of the regression equations. When various sources of data are combined to perform the analysis, significant increases in prediction occur in all cases. (Author/DEP)