A RESIDENTIAL MANPOWER TRAINING CENTER FOR ECONOMICALLY DISADVANTAGED YOUNG WOMEN WAS EXPERIENCING DIFFICULTY IN RETAINING TRAINEES DESPITE THE PROVISION OF HEALTH SERVICES, COUNSELING SERVICES, AND OTHER BENEFITS, IN ADDITION TO EDUCATION AND JOB TRAINING. SEVERAL ATTRIBUTES AND BEHAVIORS, EMERGING WITHIN THE TRAINING SETTING, WERE USED TO DEVELOP THE FOLLOWING INDEPENDENT VARIABLES RELATING TO RETENTION CRITERIA SUCH AS COMPLETION OF TRAINING OR VARIOUS STAGES OF EARLY WITHDRAWAL FROM THE PROGRAM: (1) ARITHMETIC COMPUTATION (EDUCATIONAL ACHIEVEMENT), (2) PARAGRAPH MEANING (EDUCATIONAL ACHIEVEMENT), (3) LOCUS OF CONTROL, (4) ALIENATION, (5) MANIFEST AGGRESSION, (6) AUTISTIC THINKING, AND (7) WITHDRAWAL . SUBJECTS WERE 100 FEMALE TRAINEES BETWEEN THE AGES OF 16 AND 21 SELECTED CONSECUTIVELY ON ENTRY INTO THE PROGRAM OVER A 5-MONTH PERIOD. SUBJECTS WERE LATER ASSIGNED TO ONE OF TWO GROUPS DEPENDING ON WHETHER THEY REMAINED IN THE PROGRAM BEYOND 90 DAYS. MULTIPLE STEPWISE REGRESSION ANALYSIS WAS USED TO DETERMINE THE SETS OF VARIABLES HAVING THE GREATEST CAPABILITIES FOR PREDICTION OF RETENTION IN THE PROGRAM. FOUR SETS OF VARIABLES--MANIFEST AGGRESSION, PARAGRAPH MEANING, ALIENATION, AND ARITHMETIC COMPUTATION--IN VARIOUS COMBINATIONS CONTRIBUTED SIGNIFICANTLY TO THE PREDICTION MODEL; THE REMAINING SETS DID NOT. THE MODEL SHOWS A 61 PERCENT CHANCE THAT INDIVIDUALS WHO SCORE A 10 ON THE MANIFEST AGGRESSION SCALE AND A 9.6 ON THE PARAGRAPH MEANING SCALE WILL REMAIN IN THE PROGRAM LONGER THAN 90 DAYS. THE MODEL CAN PREDICT TWO TIMES OUT OF THREE THOSE TRAINEES WHO WILL REMAIN IN THE PROGRAM AND IS APPLICABLE TO PROGRAMS IN THE HEALTH AND SOCIAL SERVICE FIELDS. REFERENCES ARE PROVIDED. (KJM)