The Mahalanobis distance model was compared with the linear discriminant function model and found to provide very similar results, even when a number of the variables were binary. A group of college freshmen were categorized into two groups: 116 "leavers," students who did not return for the second year, and 269 "returners." Data from the questionnaire Survey of Entering Freshmen was used for the predictor variables, including 34 continuous variables and 7 binary variables, including sex. Regardless of whether 7, 5, 3, or no binary variables were included, the two procedures give identical results. However, when the sample was divided into four groups, using sex as another category, minor differences in results from the two methods did appear. Adding the binary variables to the predictor variables always increased the effectiveness of the analysis, even though they violated the assumption of linearity. (Author/CTM)