Problems with using stepwise analytic methods are discussed, and better alternatives are illustrated. To make the illustrations concrete, an actual data set, involving responses of 91 medical school admissions directors to 30 variables, was used. The 30 variables involved perceptions of barriers to medical school with respect to characteristics of medical students. The propensity of researchers to apply statistical significance tests to evaluate how many steps should be implemented is considered. The problems of using statistical significance testing in conjunction with stepwise methods are elaborated, but the emphasis is on better alternatives to stepwise methods. A two-stage approach to variable selection is recommended. If variables must be eliminated, a better procedure is to compute effect sizes for every possible predictor set using readily available computer software. In the two-stage process, variable selection does not depend on the results in previous steps. The first step is to determine the desired size, "k," of the predictor variable set. The second step is to select the best predictor set of the desired size, consulting the effect size; however, a better approach would be to select the predictor set based on theory, previous empirical results, or the accessibility of variables in a given set. A 19-item list of references is included. Three tables and one graph illustrate the analysis. (Author/SLD)