There is a large literature on the estimation of frontier production functions, much of it applied to low-income agriculture. However, much of this literature largely ignores nature's role in agricultural production. Because exogenous, natural production conditions (e.g., rainfall, soil quality, pest infestation, plant disease, weed growth) are rarely uniform or symmetrically distributed within a population or a sample thereof, this omission generally leads to downward bias in producers' estimated efficiency and to biased estimates of both the parameters of the production frontier and the correlates of true technical inefficiency. Using panel data from 464 traditional rice plots in Cote d'Ivoire, we show that controlling for stochastic, exogenous, natural production conditions in estimating the production frontier significantly increases smallholder rice farmers' estimated efficiency, whether estimated using parametric, stochastic or nonparametric, nonstochastic methods. The resulting frontier parameter estimates are also more consistent with theoretical predictions than are those of a frontier estimated without controlling for exogenous production conditions. Conventional estimates of technical efficiency may then mislead policymakers' perceptions of overall efficiency levels and of the sources of such inefficiency.