There are two major problems with existing prison impact studies. First, most are poorly designed and are simply unable to determine whether measured outcomes are caused by the prison or other factors. Consequently, aggregate data are more suggestive than definitive. Second, the studies are almost always case studies, and the external validity of their findings is essentially nonexistent. Two designs are proposed, one sensitive to internal validity problems and one whose major concern is generalizability. The first design would focus on a single site and emphasize internal validity. It would be a quasiexperimental design with multiple-dimension, multiple-cause, and multiple-outcome measures and would incorporate a time series measurement approach. The results would not be generalizable, but the intensive nature of the study and the detail involved would provide an excellent picture of the dynamics of siting and opening a prison in a community. The second design would look at multiple sites and generalizability. It would be a time series, multiple-site approach using secondary data from across the United States and would focus on between-class comparison of newly sited prison communities with comparable nonprison communities. Important considerations would include multiple indicators of all concepts and close examination of measures for comparability across communities and time. It is concluded that strict attention to internal validity concerns will provide better answers to causal mechanisms and outcomes than is currently the case in prison impact studies. 15 references