IN THE IDEAL SITUATION, DIAGNOSIS LEADS TO TREATMENT, WHICH IN TURN LEADS TO OUTCOME. ALL MEMBERS OF THE DIAGNOSTIC CLASS ARE HOMOGENEOUS, ALL SEPARATE TREATMENTS ARE IDENTICAL, AND THE OUTCOME CLASSES ARE ALSO HOMOGENEOUS. FOR PURPOSES OF RATIONAL DECISIONMAKING ABOUT TREATMENT, THESE CLASSIFICATIONS ARE VALUABLE ONLY INSOFAR AS THEY LEAD TO HOMOGENEOUS TOTAL CLASSES. EXPERIENCE DOES NOT REVEAL THAT HOMOGENEITY CAN BE ACHIEVED IN THESE CATEGORIES. THERE IS AMBIGUITY NOT ONLY WITH RESPECT TO DIAGNOSIS, TREATMENT AND OUTCOME, BUT THE FACT THAT MULTIPLE INFLUENCES AFFECT OUTCOME, PRECLUDES RIGID CONTROL. IN PREDICTING RECIDIVISM, THEREFORE, STATISTICAL CATEGORIES MUST BE EMPLOYED. A REGRESSION MODEL IS USED IN WHICH EACH INDIVIDUAL IS EVALUATED ON MANY CHARACTERISTICS AND GIVEN A COMPOSITE SCORE. ON THE BASIS OF A REGRESSION EQUATION, A PREDICTION CAN BE FORMULATED FOR EACH CASE. THE PRACTICAL PROBLEMS OF SUCH EVALUATION ARE THEN EXAMINED IN LIGHT OF THE EXPERIENCE OF THE PATUXENT INSTITUTION IN MARYLAND. THIS INSTITUTION WAS ESTABLISHED FOR 'DEFECTIVE DELINQUENTS.' A JUDGE OR JURY COMMITS A PERSON TO PATUXENT AFTER A MEMBER OF THE CLINIC STAFF DIAGNOSES HIM TO BE DEFECTIVELY DELINQUENT. SENTENCES ARE INDETERMINATE AND RELEASE DEPENDS UPON SUCCESSFUL COMPLETION OF VARIOUS STEPS OF THE TREATMENT PROGRAM. IT IS CONCLUDED THAT CONSIDERABLE DIAGNOSTIC ERROR IS POSSIBLE AT EVERY STAGE OF THE PROGRAM. THE INSTITUTIONAL STAFF HAS MADE FALSE POSITIVE ERRORS TO THE EXTENT THAT OVER HALF OF THE INDIVIDUALS INCARCERATED THERE MAY NOT BE PROPERLY RETAINED ACCORDING TO THE TERMS OF THE MARYLAND STATUTE. IT IS RECOMMENDED THAT THE VIABLE TREATMENT FEATURES OF THE INSTITUTION BE RETAINED BECAUSE THEY DO SEEM TO RESULT IN REDUCED RECIDIVISM. IT IS RECOMMENDED, HOWEVER, THAT THE INDETERMINANT SENTENCE BE ABOLISHED. INSTEAD, OFFENDERS SHOULD BE SENTENCED FOR A DEFINITE PERIOD OF TIME WITH PAROLE RESERVED TO REWARD THOSE WHO SUCCESSFULLY COMPLETE VARIOUS PHASES OF THE TREATMENT PROGRAM. A BIBLIOGRAPHY IS APPENDED. (GLR)