The clinical readings produced by each team -- a psychiatrist, psychologist, and psychiatric social worker -- were studied using a combination of several methods. Seven observers recorded information on the social dynamics of clinical work in 130 diagnostic sessions. Observers noted relevant features of clinician-patient interaction and tape recorded what was said before, during, and after diagnostic interviews. Participating clinicians were subsequently interviewed about their diagnostic work. The study concludes that predictive diagnostic judgments are contingent on a complex process of social interaction, whereby clinicians construct theories about the future behavior of patients based on present reading of past records. These clinical readings are subsequently expanded, modified, and justified as clinicians selectively guide the structure of psychiatric interviews and negotiate with each other over such issues as deferential status and the practical and political consequences of a particular diagnosis. In the final diagnostic report, a multiplicity of social factors are reduced to an objective-sounding argot of individualized pathology. Current clinical methods of prediction are believed to be unjust, because they subject people to unproven and highly idiosyncratic readings of the potential for violence. Several proposals for reforming the prediction process are discussed, with the aim of reducing the arbitrary and capricious nature of clinical assessments. Thirty-one references are listed.