Impact evaluations have exploded in recent years in development economics. While they have internal validity, they have no theoretical external validity. Many conducting impact evaluations would now say that their main purpose is not to gauge the effects of a particular program but to understand the causal mechanisms behind the estimated outcomes. Yet on top of what we can learn from an impact evaluation along those lines, every study also has a certain amount of de facto external validity which is important both for policy makers and for researchers seeking to make an original contribution. This paper is the first to estimate the de facto external validity of impact evaluations of many kinds of international development programs, using a unique data set of impact evaluation results. A theory is developed about how characteristics of a particular impact evaluation may affect the generalizability of its results to different settings, and the data are used to test this theory.