The first methodological problem in the study of predicting dangerous behavior is that there is no systematic way to test predictions. The only completely accurate means is to make the assessment that a given subject is dangerous and would probably reoffend, release the offender, and observe his subsequent behavior. Obviously, this method is totally impractical. Second, thus far no structured or projective test scale has been derived which, when used alone, will predict violence in the individual case in a satisfactory manner. Tests considered for this purpose include the Rorschach, the Thematic Perception Test, the Minnesota Multiphasic Personality Inventory, and others. Third, although everyone supposedly knows what violence is, there is no generally agreed upon semantic or operational definition of the term. More people are killed, injured, and maimed annually by drunken drivers than by any other group of offenders. Yet drunken drivers are not considered to be included in the implicit concept of dangerousness employed by many who make such assessments. Fourth, rigorous studies of the psychology of human violence are difficult to perform. There is some agreement on the personality factors that interact to determine whether or not an individual will engage in violent behavior. However, situational factors vary and may operate to inhibit or deter aggressive behavior. Thus, clinical assessment involves taking small samples of behavior to estimate future behavior. Just as in political polling, such a process inevitably involves error. The limitations of assessment techniques make it impossible to predict with perfect accuracy. Three tables and seven references are provided.