Randomized experiments in development, as practiced by Duflo, Banerjee, Mullainathan, Kremer, etc., have become particularly prominent. Nonetheless, they are often criticized as being atheoretical, as having inherent biases, and of having little external validity. The authors point out that these worries are often well-founded. However, carefully done random trials can be used to test theory or structured in a way that allows identification in structural models (therefore complementing non-randomized work). Worries about “scaling up” – finding that funding private school tuition in a village improves outcomes does not imply that funding private school nationwide will do the same, since the existing private schools do not have enough capacity – can sometimes be handled by variation within the experiment, such as funding a greater percentage of villages in the catchment of one private school than in another. Worries that an experiment has effects in Kenya means very little for whether the experiment will have similar effects in India imply only that experiments ought be replicated, and large foundations such as Ford are funding such replication even if economics journals do not themselves reward replication. The paper is in the new Annual Review of Economics, and as such is quite general, providing citations to a number of interesting papers on education, health, and microfinance research.