In the context of this debate, we review research concerning the prediction of violence, and conclude that because such predictions are of low accuracy, they are only modestly useful for public policy formulation or for individual decision-making. Because we believe prediction, and utilitarian goals, to be important, this paper focuses on two issues that have potential for increasing the accuracy with which predictions may be made. One is the measurement of the seriousness of crime and ways to improve it. Second, we introduce the concept of societal stakes and suggest that this must be assessed as well. Finally, we propose a model that may be useful for lessening the conflict between retributive and utilitarian perspectives. 3 footnotes; 83 references. (Author abstract)