The principles are that (1) dangerousness must be disaggregated, (2) an array of risk factors in multiple domains must be chosen, (3) harm must be scaled in terms of seriousness and assessed with multiple measures, (4) risk must be considered a probability estimate that changes over time and context, (5) priority must be given to actuarial research that establishes a relationship between risk factors and harm, (6) retrospective samples are needed, and (7) managing as well as assessing risk must be a research goal. These principles were derived from decision theory and public health. A field trial of the provisional research design was conducted from 1988 to 1990. The results were used to develop the research design for a full-scale trial. A sample size of approximately 1,000 mental patients from three civil mental hospitals is anticipated. If the study is successful in its efforts to establish robust markers of violence risk, future research might usefully address optimum methods of educating clinicians in the use of these markers. Tables and 34 references