TWO REASONS ACCOUNT FOR THE FAILURE TO PREDICT PAROLE OUTCOMES. FIRST, THE PREVALENT USE OF PRISON FILES AS A DATA SOURCE LIMITS THE KIND OF RESEARCH QUESTIONS THAT CAN BE ASKED. SECOND, THE NONTHEORETICAL NATURE OF PAROLE PREDICTION RESEARCH HAS PREVENTED THE SYSTEMATIC ACCUMULATION OF KNOWLEDGE PERTAINING TO PREDICTION. WHILE THE SUCCESS OF PAROLE PREDICTION RESEARCH HAS BEEN LIMITED, THE PRACTICAL AND THEORETICAL POTENTIAL OF SUCH EFFORTS HAS SUSTAINED INTEREST OVER A LONG PERIOD OF TIME. IF THAT POTENTIAL IS TO BE REALIZED, TWO TYPES OF CHANGES MUST BE MADE. PRISON ADMINISTRATORS MUST RECOGNIZE THE NEED FOR BASIC RESEARCH IN THE CORRECTIONAL SETTING AND MAKE RELEVANT TYPES OF DATA AVAILABLE. IN ADDITION, PREDICTION RESEARCH MUST BE GUIDED BY THEORY SO THAT AVAILABLE RESOURCES CAN BE USED MOST EFFICIENTLY. PAROLE PREDICTION RESEARCH NEEDS TO CONFORM TO THE BASIC REQUIREMENTS OF SCIENTIFIC INVESTIGATION, WHICH INCLUDE THE SELECTION OF VARIABLES ON THE BASIS OF THEIR THEORETICAL RELEVANCE, NOT THEIR AVAILABILITY IN PRISON FILES. REFERENCES TO SPECIFIC STUDIES FROM 1923 THROUGH THE 1960'S ARE INCLUDED. (AUTHOR ABSTRACT MODIFIED--LKM)