A review of the history of analyses of "what works" concludes that the dearth of ongoing, responsible research in correctional programming has resulted in two failures: the failure to routinize program evaluation and the inadequacies of the methodologies used by researchers as reported in the published literature. In addition to addressing these two failures, the search for "what works" must also recognize that even the most current and sophisticated scientific methods and results provide provisional knowledge that is subject to the inevitable limitations of any search for truth. Further, evaluations seldom show a program to be totally effective in achieving intended results. A program will most often be effective in significantly reducing criminal behavior for some participants but not for others. Such a circumstance requires a value judgment regarding whether the resources invested are worth the partial outcomes. Sherman et al. developed a useful typology for classifying programs based on their evaluation findings. One program category is labeled "What Works," which are programs that are reasonably certain to achieve their primary goals; a second category of programs are classified as "What Doesn't Work," which are programs that are reasonably certain, from available evidence, to fail in meeting their primary goals; the third category of programs is called "What's Promising," which are programs for which the level of certainty from available evidence is too low to support generalizable conclusions, but for which there is some empirical basis for predicting that further research could support such conclusions; the fourth category of programs is called, "What's Unknown," which pertains to any program not classified in one of the other three categories. In its concluding sections, this article discusses how correctional administrators engage in decision-making about a program based on its classification in one of the aforementioned categories. The author advises that successful evaluation will not happen automatically; it requires leadership by the administrator, a commitment to evaluation research that flows from explicit goals, and a willingness to identify and accept public policy as an inevitable aspect of responsive and responsible decision-making. 50 references