Over the last 15 years, development economists have carefully accumulated rigorous evidence about what works and what does not in promoting health in poor countries. While each individual evaluation tests specific questions or sets of questions in specific contexts, the large number of studies now means that it is possible to draw more general conclusions. In addition, randomized evaluations are increasingly being designed to test fundamental questions about how people behave and thus generate lessons that are relevant for the design of different types of programs. In this seminar, Michael Kremer will discuss a new research paper co-authored with Rachel Glennerster, Lessons from Randomized Evaluations for Improving Health in Developing Countries, which summarizes lessons from the growing body of randomized evaluations of health programs in developing countries. The paper finds considerable evidence that consumers do not always invest optimally in health. In particular, consumers underinvest in cost-effective products for prevention and non-acute care of communicable disease and are very sensitive to the price and convenience of these products. This underinvestment does not simply reflect a lack of information of the benefits of preventative health. While this suggests the need for government intervention, many government health systems perform poorly and there is little accountability and few incentives for health care providers. Of the approaches designed to improve accountability, community or nongovernmental monitoring has had mixed results but district-level contracting has been quite successful. Many programs can improve health without excessive reliance on dysfunctional health delivery systems—delivering health products through schools for example, or improving health through water treatment.