The study reported on here is the first benefit-cost analysis of a controlled (random assignment) experiment in the mental health field. It compares, in terms of an unusually wide variety of tangible and intangible forms of benefits and costs, a traditional, hospital-based approach to treating the mentally ill with a nontraditional community-based approach. The research reported here supports the hypothesis that hospitalization of the mentally ill is, except for emergencies, less effective than community-based treatment of approximately equal cost. The research also confirms the hypothesis that the forms taken by the social costs of alternative programs can be so different that it is easy to mistake a change in the form of costs for a change in their level. Finally, this study highlights the fact that benefit-cost analysis, despite advances at both the conceptual and empirical levels, remains a mixture of science and art.