In a large-scale agricultural model a sizeable amount of low level data undergoes several stages of processing. In designing a model, uncertainty and complexity will restrict the generality of computer-implemented stages and much of the processing will be manual. The resulting model is likely to be inflexible to changes or else require as long again to modify and rerun as to design another model in order to evaluate another policy. By recognizing the necessary constraints imposed by mass data handling, the proportion of time spent on data problems could be reduced. Emphasis should be placed on defining beforehand as explicity as possible an acceptable methodology and what the specific uses of the model will be.