Methods for ordering and selecting variables for discriminant analysis in multiple group comparison or group prediction studies include: univariate Fs, stepwise analysis, learning discriminant function (LDF) variable correlations, communalities, LDF standardized coefficients, and weighted standardized coefficients. Five indices based on distance, association, and classification accuracy, evaluate the sense of discriminatory power in which a subset of a given size determined by one method, would be better than a subset of the same size determined by another method. To illustrate the use of the methods and indexes, thirteen variables for assigning 153 college freshmen to one of three French courses were rank ordered according to all six methods. The five indexes were computed for subsets of each of ten sizes determined by each method. The relative effectiveness of the variable selection methods for yielding good subsets depended on the index used. The methods agreed closely for the distance and association indexes for all subset sizes (concordance coefficient .91); but they were inconsistent for the classification index (concordance coefficient .17). The combination of variable selection method and purpose of discriminant analysis research, and the relationship between selection method and structure of data need further study. (CP)