This paper suggests practical methods for assessing policy research programs, both ex post and ex ante. Measuring the benefits of policy research is difficult: the path of causation between research and policy change is nearly always uncertain; multiple factors influence any particular policy change; policies are diverse in nature as are their intended and actual effects; and some effects of policy research are not priced in the market. Many of the benefits of changes in policy stem from the reduced cost of welfare-improving institutional change. Economic surplus analysis can be used to assess such changes. In some cases, Bayesian decision theory may be helpful in evaluating policy research, although it is usually difficult to obtain estimates of the probability distributions a decision-maker has before the research becomes available. Subjective estimates of parameters and some measure of their degree of uncertainty, are likely to be needed for an economic surplus model. The paper suggests a set of steps for policy research evaluation. It is applied to two cases: an evaluation of pesticide policy research in Brazil, and an evaluation of policies affecting deforestation in Indonesia.