Quite different R&D lag structures predominate in studies of agricultural R&D compared with studies of R&D in other industries, and compared with studies of economic growth more broadly. Here we compare the main models and their implications using long-run data for U.S. agriculture. We reject the models predominantly used in studies of economic growth and industrial R&D both on prior grounds and using various statistical tests. The preferred model is a 50-year gamma lag distribution model. The estimated elasticity of MFP with respect to the knowledge stock is 0.28 and the implied marginal benefit-cost ratio is 23:1.