International organizations (IOs) have long been a central focus of scholarship in international relations, yet we know remarkably little about their performance. This article offers an explanation for differences in the performance of IOs and tests it using the first quantitative data set on the topic. I argue that the primary obstacle to effective institutional performance is not deviant behavior by IO officials—as conventional “rogue-agency” analyses suggest—but the propensity of states to use IOs to promote narrow national interests rather than broader organizational objectives. IOs that enjoy policy autonomy vis-à-vis states will thus exhibit higher levels of performance. However, in the international context policy autonomy cannot be guaranteed by institutional design. Instead, it is a function of (1) the existence of (certain types of) institutionalized alliances between IOs and actors above and below the state; and (2) the technical complexity of IO activities. I provide empirical evidence for the argument by constructing and analyzing a cross-sectional data set on IO performance—based in part on a new wave of official government evaluations of IOs and in part on an original survey of IO staff—and conducting a comparative case study in the realm of global food security.