The researchers followed 618 male offenders who had been treated in a maximum-security facility or were assessed there prior to imprisonment. Over an average time at risk of 81.5 months, 31 percent of these men committed a new violent (including sexual) offense against persons. Multivariate statistical procedures were used to select 12 variables for use in a prediction instrument. The violent recidivism rate for each value or range of values for each variable was determined. For every difference of more than 50 percent from the mean overall violent recidivism rate, a weighting of plus or minus 1 was added, depending on whether the value was associated with an increase or a decrease in recidivism from the mean. Each subject could then be scored on each variable, and the scores could be added to form a single risk score for each offender. There was a linear relationship between risk score and likelihood of violent recidivism. The proportion of violent recidivists correctly identified was .41, and specificity (the proportion of successes correctly identified) was .88. The results show that dangerousness (the disposition to commit violent offenses) is linearly related to risk score. The author advises that the accuracy of the assessment of dangerousness achieved with the prediction instrument is dependent on the assessed person being a serious offender. The type of followup research leading to the development of the prediction instrument is ill suited to discovering variables related to a person becoming or not becoming a violent offender in the first place. The discussion focuses on factors related to the psychopathy category of offenders, since most violent offenders come from this category. 39 references and 1 figure