With whole-farm models, it is usually hard to know how representative the models are of reality. No rigorous validation procedure seems to have been proposed. There has always been a dilemma between descriptive power and tractability. The outcome has been to resort to a confused mix of expert opinion and normative attitude. Discrepancies between model output and farmers' behaviour have usually been interpreted in against farmers, by considering them to be in some way sub-optimal or insufficiently informed. In traditional systems where the rationale of production behaviour can vary widely, it may pay to first understand their rationale before applying normative principles or recommending new technologies. We use a case study in Mexico to illustrate a procedure allowing for such an assessment of our understanding. The procedure, though based on the use of linear programming, is independent of the modelling technique. (C) 1997 Published by Elsevier Science Ltd.