Protected areas are leading tools in efforts to slow global species loss and appear also to have a role in climate change policy. Understanding their impacts on deforestation informs environmental policies. We review several approaches to evaluating protection’s impact on deforestation, given three hurdles to empirical evaluation, and note that “matching” techniques from economic impact evaluation address those hurdles. The central hurdle derives from the fact that protected areas are distributed nonrandomly across landscapes. Nonrandom location can be intentional, and for good reasons, including biological and political ones. Yet even so, when protected areas are biased in their locations toward less-threatened areas, many methods for impact evaluation will overestimate protection’s effect. The use of matching techniques allows one to control for known landscape biases when inferring the impact of protection. Applications of matching have revealed considerably lower impact estimates of forest protection than produced by other methods. A reduction in the estimated impact from existing parks does not suggest, however, that protection is unable to lower clearing. Rather, it indicates the importance of variation across locations in how much impact protection could possibly have on rates of deforestation. Matching, then, bundles improved estimates of the average impact of protection with guidance on where new parks’ impacts will be highest. While many factors will determine where new protected areas will be sited in the future, we claim that the variation across space in protection’s impact on deforestation rates should inform site choice.