TO DATE, EFFORTS TO PREDICT DANGEROUSNESS AMONG MENTALLY DISORDERED OFFENDERS HAVE BEEN HAMPERED BY MISPLACED RELIANCE ON THE USE OF THE SIMPLE DANGEROUS/NOT DANGEROUS CLASSIFICATIONS, INADEQUATE DEFINITIONS OF DIFFERENT TYPES OF DANGEROUS BEHAVIOR, AND CONFOUNDING OF THE POSSIBLE CUES USEFUL IN THE PREDICTION OF VIOLENCE. DATA SHOW THAT BEHAVIORS CLINICIANS SOMETIMES CALL 'DANGEROUS' DEPEND ON THE QUALITY OF THE DATA-RECORDING SYSTEM USED. STUDIES BASED ON EXAMINATION OF PSYCHIATRISTS' LETTERS POST HOC MAY, FOR EXAMPLE, CONTRIBUTE TO THE PREVALENCE OF OVERPREDICTION. SIMPLE DANGEROUS/NOT DANGEROUS DICHOTOMIES SHOULD BE AVOIDED. THE CLINICIAN SHOULD DEVELOP A RANGE OF POSSIBLE SCORES WITHIN WHICH TO EXPRESS AN OPINION. DATA HAVE SHOWN HOW THE USE OF A FOUR-STEP SCALE ALTERS THE PREDICTION PROBLEM, AND OTHER PRELIMINARY DATA TAKEN FROM PSYCHIATRIC INTERVIEWS HAVE ILLUSTRATED HOW ATTENTION TO DEFINING DIFFERENT TYPES OF DANGEROUS BEHAVIOR MAY ENABLE RESEARCHERS TO SEE PATTERNS AND CONFIGURATIONS NOT OTHERWISE APPARENT. DATA FROM THIS STUDY INDICATE THAT CLINICAL OPINION IS INFLUENCED BY BACKGROUND VARIABLES SUCH AS CURRENT CHARGE AND PREVIOUS EXPERIENCE. ALTHOUGH THERE MAY BE GOOD REASON TO BASE PREDICTIONS PARTLY ON SUCH FACTORS, UNDER IDEAL ASSESSMENT CONDITIONS, A PREDICTION EQUATION SHOULD BE DEVELOPED IN WHICH EACH ELEMENT (INCLUDING THE CRUCIALLY IMPORTANT INTERVIEW) IS EVALUATED SEPARATELY. GREATER EMPHASIS SHOULD BE GIVEN TO DEFINING, DESCRIBING, AND RECORDING WHAT OCCURS BETWEEN PATIENT AND PSYCHIATRIST DURING THE COURSE OF THE ASSESSMENT INTERVIEW. TABULAR DATA AND REFERENCES ARE PROVIDED. (AUTHOR ABSTRACT MODIFIED --RCB)