THREE FACTORS SHOULD BE CONSIDERED IN SELECTING MEASURES OF DELINQUENCY: THE WAY IN WHICH THE DATA COLLECTED WILL BE USED (TO MEASURE IMPACT OF PREVENTION PROGRAMS ON DELINQUENCY RATES OR TO DEVELOP WAYS OF REDUCING THE FREQUENCY OF DELINQUENT BEHAVIOR AMONG A PARTICULAR POPULATION OF YOUTHS), THE EASE WITH WHICH THE DATA CAN BE COLLECTED, AND THE RELIABILITY OF THE DATA. OFFICIAL STATISTICS CAN BE OBTAINED FROM LAW ENFORCEMENT AGENCIES, AND SELF-REPORT DATA CAN BE OBTAINED CONFIDENTIALLY FROM CLIENTS INVOLVED IN DELINQUENCY PREVENTION PROGRAMS. THE RELIABILITY OF THESE TWO MEASURES IS ASSESSED. A HISTORICAL REVIEW OF PREDICTION RESEARCH IS PRESENTED. PREDICTOR VARIABLES THAT WERE INVESTIGATED MAY BE CLASSIFIED AS SOCIOLOGICAL AND BACKGROUND, PSYCHOLOGICAL OR PERSONALITY, AND BEHAVIORAL. THE MOST EFFICIENT PREDICTION METHOD IS ONE WHICH INCORPORATES RELIABLE INDICATORS FROM EACH OF THESE THREE THEORETICAL PERSPECTIVES. INVESTIGATORS SHOULD NOT ATTEMPT TO IDENTIFY A SINGLE CAUSE OF DELINQUENT BEHAVIOR BUT RATHER SEARCH FOR A SERIES OF VARIABLES WHICH INTERACT TO CAUSE DELINQUENCY. THE PRIMARY ARGUMENTS AGAINST THE USE OF PREDICTION INVOLVE OVERPREDICTION AND LABELING YOUTH AS PREDELINQUENT. ISSUES TO BE CONSIDERED IN DEVELOPING PREDICTION METHODOLOGIES ARE NOTED. THE MOST EFFICIENT METHOD OF DEVELOPING A PREDICTION SCALE INVOLVES THE USE OF THEORETICAL CONCEPTS OF DELINQUENCY CAUSATION IDENTIFIED BY PATH ANALYSIS, WHICH MEASURES HIERARCHICAL INTERRELATIONSHIPS AMONG VARIABLES AND WEIGHS THEM STATISTICALLY THROUGH MULTIPLE REGRESSIONS. SINCE A PREDICTION SCALE COMPRISES THOSE VARIABLES MOST LIKELY TO CAUSE OR RESULT IN DELINQUENT BEHAVIOR, IT CAN BE USED TO IDENTIFY AREAS IN WHICH INTERVENTION OR REMEDIAL SERVICES SHOULD BE PROVIDED. PREVENTION WOULD INVOLVE ASSURING THE AVAILABILITY OF SERVICES TO ALLEVIATE EXISTING PROBLEMS AND PERIODIC ASSESSMENT OF DELINQUENT-PRONE YOUTHS. MORE INTENSIVE PREVENTION SERVICES COULD BE PROVIDED TO YOUTHS FAILING TO SHOW IMPROVEMENTS ON PREDICTOR VARIABLES. (MGB)