This essay is a critique of research evaluation research. Considerable evidence exists that agricultural research conducted during the era when projects were chosen by diffuse selection systems yielded extraordinarily high returns. It is not obvious that the formalized, quantitative, and typically centralized selection models can be expected to produce higher contemporary returns than the decentralized informal mechanisms. All ex ante evaluations are intrinsically subjective, regardless of technique used to generate the evaluation. The extreme uncertainty surrounding the nonrepetitive new-knowledge production function further limits the potential of the sophisticated selection procedures. Perhaps of greatest importance, however, are the high costs imposed by these procedures in terms of scientists' time, morale, and "artistic" research tool atrophication.