EXCLUSION OF DRUG OFFENDERS REFERRED TO IT BY THE COURTS CAUSED CONFLICTS BETWEEN THE CALIFORNIA REHABILITATION CENTER AND JUDGES, WHO RESENTED HAVING THEIR DISPOSITIONS QUESTIONED, BUT APPELLATE COURTS ALWAYS SUSTAINED THE CENTER'S DECISIONS. HOWEVER, SINCE 1974, THE CALIFORNIA REHABILITATION CENTER BEGAN TO ACCEPT AN INCREASING NUMBER OF MARGINAL CASES WHICH WOULD HAVE PREVIOUSLY BEEN EXCLUDED ON THE EXPECTATION THAT THEY WOULD BE UNTREATABLE OR UNCONTROLLABLE. THE EFFICACY OF EXPANDING THE CALIFORNIA REHABILITATION CENTER'S ADMISSION POLICY WAS TESTED BY COMPARING THE POST-SELECTION BEHAVIOR OF A SAMPLE OF MARGINALLY RETAINED CASES TO THAT OF TWO SAMPLES OF MORE CLEARLY ACCEPTABLE CASES. THE ENTIRE STUDY POPULATION COMPRISED 350 ADDICTS REPRESENTATIVE OF ADMINISTRATIVE CLASSIFICATION ACTIONS TAKEN DURING THE LAST QUARTER OF 1974. THE MARGINAL SAMPLE WAS FOUND TO BE SLIGHTLY MORE DELINQUENT IN THE INSTITUTION (CALIFORNIA REHABILITATION CENTER), BUT THEIR DELINQUENCY INVOLVED MINOR INFRACTIONS RATHER THAN SERIOUS OFFENSES (E.G., NARCOTICS USE OR ASSAULTS). THERE WAS NO DIFFERENCE AMONG THE SAMPLES ON COMPREHENSIVE MEASURES OF OUTPATIENT ADJUSTMENT IN THE COMMUNITY (E.G., ARRESTS, CONVICTIONS, NARCOTICS USE, ABSCONDING). THE GROUP EXCLUDED DURING THE SELECTION QUARTER AND TRACKED THROUGH SUBSEQUENT ADJUDICATION WAS DISCOVERED TO HAVE RECEIVED A LOCAL COMMUNITY LEVEL DISPOSITION (JAIL AND/OR PROBATION). THEREFORE, IT SPENT ABOUT THE SAME TIME IN CUSTODY AS DID THE MARGINAL SAMPLE, WHICH REMAINED IN THE CALIFORNIA REHABILITATION CENTER. BASED ON THE SURVEY FINDINGS, THIS STUDY CONCLUDES THAT EXPANDING THE MARGINAL RETENTION POLICY DID NOT LOWER PROGRAM PERFORMANCE OR INCREASE THE LEVEL OF DELINQUENT BEHAVIOR IN A PRACTICAL SENSE. NUMEROUS STATISTICAL AND TABULAR DATA ARE PROVIDED IN THE TEXT AND A LIST OF BIBLIOGRAPHIC REFERENCES IS APPENDED. TWO APPENDIXES GIVE (1) THE CIVIL ADDICT PROGRAM'S GUIDELINES AND CRITERIA FOR THOSE ELIGIBLE AND (2) TABLES ON THE CALIFORNIA REHABILITATION CENTER'S POPULATION CLASSIFIED ACCORDING TO DESCRIPTIVE VARIABLES.