POSTPAROLE OUTCOMES WERE KNOWN FOR ALL CASES. THE STUDY SELECTED 40 PREDICTIVE VARIABLES (DEMOGRAPHIC, SOCIAL, FAMILY, PERSONAL, AND PSYCHOLOGICAL MEASURES) FROM THE CASE HISTORIES. THE SAMPLE WAS THEN DIVIDED RANDOMLY INTO TWO GROUPS. THREE PAROLE PREDICTION ANALYSES WERE RUN ON EACH GROUP, ONE USING EACH OF THE MULTIVARIATE TECHNIQUES. AFTER THE PREDICTORS WERE SELECTED FOR EACH SAMPLE, THEY WERE CROSS VALIDATED WITH THE OTHER GROUP. IN ALL, A TOTAL OF SIX ANALYSES WERE RUN. THE RELIABILITY OF EACH METHOD WAS ASSESSED ON THE BASIS OF THE DEGREE OF DISCRIMINABILITY ASSOCIATED WITH EACH VARIABLE, THE ACCURACY OF PREDICTION FOR EACH JUVENILE CASE, AND THE DETERMINATION OF WEIGHTINGS OF EACH VARIABLE IN THE PREDICTION PROBLEM. THE PREDICTOR SETS SELECTED BY STEPWISE REGRESSION AND FACTOR ANALYSIS SHOWED LITTLE OVERLAP. BY CONTRAST, THE DISCRIMINANT FUNCTIONS METHOD PROVIDED THE MOST CONSISTENT SELECTION WITH 7 OF THE 10 VARIABLES THE SAME. (TABLES DETAIL THE RESULTS OF THESE VARIOUS ANALYSES.) DISCRIMINANT FUNCTIONS ALSO SHOWED THE GREATEST PREDICTIVE EFFICIENCY WITH SIGNIFICANTLY FEWER MISCLASSIFICATIONS (AN 11 PERCENT TOTAL ERROR RATE COMPARED WITH 20.9 PERCENT FOR STEPWISE MULTIPLE REGRESSION AND 29 PERCENT FOR FACTOR ANALYSIS). RESULTS ARE DISCUSSED IN TERMS OF THE SHRINKAGE PROBLEMS OF REGRESSION ANALYSES AND THE HOMOGENEITY OF VARIANCE OF FACTOR ANALYSIS. REFERENCES ARE APPENDED. (GLR)