CLINICAL TECHNIQUES RELY ON THE PROFESSIONAL EXPERTISE AND THEORETICAL ORIENTATION OF THE PREDICTOR. THE CAMBRIDGE-SOMERVILLE YOUTH STUDY CAUGHT BOYS WHO WERE IDENTIFIED AS POTENTIAL DELINQUENTS TO TEST A PARTICULAR PREVENTION STRATEGY. AS PART OF THE PROGRAM, THREE JUDGES WERE ASKED TO PREDICT DELINQUENT BEHAVIOR BASED ON AN EXAMINATION OF THE BOYS' RECORDS CONTAINING INTERVIEWS WITH PARENTS AND TEACHERS AND A MEDICAL REPORT. THE JUDGES USED NUMERICAL RATINGS, BUT ALSO KEPT RECORDS ON THE REASONS FOR THEIR PREDICTIONS. THIS STUDY USED CASES OF 255 BOYS TREATED BETWEEN 1939 AND 1943 WITH INFORMATION ON VIOLATIONS OF THE LAW WHEN ALL SUBJECTS WERE 30-YEARS OLD. THE JUDGES HAD PREDICTED THAT APPROXIMATELY THREE-FIFTHS OF THE CASES WOULD BECOME DELINQUENTS, WHEREAS LESS THAN ONE-THIRD ACTUALLY HAD PROBLEMS WITH THE LAW. THE FACTOR THAT ACCOUNTS MOST FOR THIS ERROR IS THE JUDGES' BROAD DEFINITION OF DIFFICULTY WHICH INCLUDED SOCIAL MALADJUSTMENT AS WELL AS OFFICIAL CONTACT WITH THE POLICE. THEY MAY HAVE ALSO OVERESTIMATED THE LONG-RANGE CONSEQUENCES OF CHARACTERISTICS REGARDED AS CRIMINOGENIC AND APPLIED TOO MANY THEORIES OF DELINQUENT BEHAVIOR. STATISTICAL TECHNIQUES WERE USED TO EVALUATE THE CORRELATIONS AMONG PARTICULAR CHARACTERISTICS, THE JUDGE'S PREDICTIONS, AND ACCURACY. ALTHOUGH THE CAMBRIDGE-SOMERVILLE STUDY DOES NOT VALIDATE THE CLINICAL METHOD, IT DOES ILLUSTRATE ITS VALUE IN RECOGNIZING INTERACTIVE RELATIONSHIPS AMONG VARIABLES. THIS METHOD COULD ACHIEVE GREATER CONSISTENCY AND ACCURACY IF COMBINED WITH MODERN STATISTICAL TECHNIQUES. FOOTNOTES ARE PROVIDED. FOR RELATED ARTICLES, SEE NCJ-64125. (AUTHOR ABSTRACT MODIFIED--MJM)