A total of 294 studies of returns to agricultural R&D (including extension) were compiled and these studies provide 1,858 separate estimates of rates of return. This includes some extreme values, which are implausible. When the highest and lowest 2.5 percent of the rates of return were set aside, the estimated annual rates of return averaged 73 percent overall–88 percent for research only, 45 percent for research and extension, and 79 percent for extension only. But these averages reveal little meaningful information from a large and diverse body of literature, which provides rate-of-return estimates that are often not directly comparable. The purpose of this study was to go behind the averages, and try to account for the sources of differences, in a meta-analysis of the studies of returns to agricultural R&D. The results conform with the theory and prior beliefs in many ways. Several features of the methods used by research evaluators matter, in particular assumptions about lag lengths and the nature of the research-induced supply shift.