In general, reported rates of return to agricultural R&D are high, but questions have been raised about upward biases in the evidence. Among the reasons for this bias, insufficient attention to attribution aspects. Matching of research benefits and costs is a pervasive problem, the magnitude of which is illustrated here with new evidence for Brazil. Over the period 1981 to 2003, varietal improvements in upland rice, edible beans, and soybeans yielded benefits attributable to research of $14.8 billion in present value (1999 prices) terms; 6.1 percent of the corresponding value of crop output. If all of those benefits were attributed to Embrapa, a public research corporation accounting for more than half Brazil's agricultural R&D spending, the benefit-cost ratio would be 78:1. If a geometric attribution rule based on genetic histories is used in conjunction with quantitative evidence on the extent of research collaborations to account for the innovative effort of others, the ratio drops substantially to 16:1. The sources of these gains vary markedly among crops and over time, making it hard to generalize about the international and institutional origins of varietal innovations in Brazilian agriculture during the past several decades.