A method of scaling qualitative data is proposed, which is useful for discriminant analysis with more than two populations. In this method, values are assigned to categories within each variable so as to maximize the Rao's V statistic. Since the method proposed here makes much use of the information about a variable as a set of categories, the usual procedure of variable selection can be easily applicable to discrete variables by employing the maximized V statistic as a measure of discrimination. As an example, the method is applied to medical data of patients suffering from four kinds of diseases. The analysis shows that the use of the V statistic is efficient as a criterion for the selection of discrete predictor variables, although this procedure of selection is somewhat heuristic and done in a backward elimination manner. Furthermore, a comparison of the proposed method with the Hayashi's quantification theory of the second type is discussed.