THE DISCUSSION, WHICH INCLUDES PARTICULAR REFERENCE TO THE CAMBRIDGE (ENGLAND) STUDY IN DELINQUENT DEVELOPMENT, DIFFERENTIATES BETWEEN THREE TYPES OF STUDIES: THOSE WHICH, LIKE THE CAMBRIDGE STUDY, ATTEMPT TO PREDICT WHICH PERSONS IN A NONDELIQUENT GROUP WILL BECOME DELINQUENTS IN THE FUTURE; THOSE WHICH AIM TO PREDICT WHICH PERSONS IN A DELIQUENT GROUP WILL BE RECONVICTED IN THE FUTURE; AND THOSE WHICH AIM TO PREDICT FUTURE RATES OF DELINQUENCY OR INCARCERATION IN SOME POPULATION. TWO CLASSIFICATIONS OF METHODOLOGY--PREDICTIVE AND PROSPECTIVE AND PREDICTIVE AND RETROSPECTIVE--ARE ALSO DEFINED, AND IT IS SUGGESTED THAT THE FORMER, IN WHICH PREDICTIONS ARE MADE BEFORE THE VALUES OF CRITERION VARIABLES ARE KNOWN, IS THE MORE SOUND. REASONS FOR CONDUCTING PREDICTION RESEARCH ARE DISCUSSED, INCLUDING THE COMPARISON OF THEORIES ABOUT CAUSES OF CRIME AND DELINQUENCY, AND THE PROVISION OF INFORMATION ABOUT THE LIKELIHOOD OF RECIDIVISM AND ABOUT THE EFFECTIVENESS OF DIFFERENT PENAL TREATMENTS. PROBLEMS ASSOCIATED WITH ATTEMPTS TO GAIN SUCH INFORMATION THROUGH PREDICTION RESEARCH ARE CITED, AND THE LACK OF THEORETICAL UNDERPINNING IN MOST STUDIES IS NOTED. THE CHOICE OF CRITERION AND PREDICTOR VARIABLES AND OF STATISTICAL TECHNIQUES FOR MEASURING THE EFFECTIVENESS OF PREDICTION IS DISCUSSED. STATISTICAL METHODS OF SELECTING AND COMBINING PREDICTORS ARE COMPARED. IT IS CONCLUDED THAT IN SMALLER SAMPLES LITTLE IS GAINED BY USING THE MORE SOPHISTICATED STATISTICAL METHODS. THE GLUECK METHOD APPEARS TO BE THE MOST SUCCESSFUL. THERE IS ALSO EVIDENCE TO SUGGEST THAT THE RELATIVE UTILITY OF THE MORE SOPHISTICATED METHODS DOES NOT INCREASE AS SAMPLE SIZE INCREASES. THE FAILURE OF SOPHISTICATED STATISTICAL TECHNIQUES TO OFFER BETTER RESULTS THAN CRUDER METHODS MAY REFLECT THE FACT THAT MOST CRIMINOLOGICAL DATA VIOLATE THE ASSUMPTIONS UPON WHICH THE SOPHISTICATED TECHNIQUES ARE BASED. ONE APPROACH MIGHT BE TO IMPROVE THE QUALITY OF THE DATA OR TO USE TECHNIQUES THAT DO NOT DEPEND ON ASSUMPTIONS THAT ARE VIOLATED. IT APPEARS, HOWEVER, THAT THERE IS A LIMIT TO THE EXTENT TO WHICH CRIMINOLOGICAL CRITERIA CAN BE PREDICTED. THE VALUE OF SELECTING AND COMBINING PREDICTORS INTO A PREDICTION INSTRUMENT IS QUESTIONED. IT IS SUGGESTED THAT THE AIM OF CRIMINOLOGICAL PREDICTION STUDIES SHOULD BE TO TEST CRIMINOLOGICAL THEORIES, RATHER THAN TO DERIVE PREDICTION INSTRUMENTS. SUPPORTING DATA FROM THE CAMBRIDGE STUDY AND A LIST OF REFERENCES ARE INCLUDED. (LKM) (LKM)