The purpose of this paper is to estimate a bias‐free agricultural production function with a view to examining efficiency differences among small rice farmers. Simultaneous equation bias is avoided if we assume that farmers maximise expected profits; specification bias, which commonly occurs when a management input is omitted from such functions, is circumvented by introducing farm‐specific dummy variables into a combined cross‐sectional and time‐series data set. Applying this model to data for 32 Philippine rice farms between 1970 and 1979, rather small production‐elasticities are obtained for the conventional inputs and an efficiency ranking of the farms is presented. Second stage analysis shows that differences in soil type, land tenure, education and access to credit are important factors explaining these efficiency differences.