The effect of model elaboration upon predictive accuracy has not been adequately studied for simulation models of crop production. A set of eight simple dynamic simulation models of alfalfa (Medicago sativaL.) production were formulated to determine the relative contribution of six factors to model accuracy. The six factors were production potential (maximum yields in each harvest), heat summation, nonlinear temperature effects, soil water budgets, regrowth potential (cutting management and root reserves), and a general soil production factor based on available soil water in the root zone at field capacity. The soils considered were a Glossoboric Hapludalf, an Aqueptic Fragiudalph, and a Dystric Eutrochrept. The factors were compared individually and in selected combinations by regressing field observations on model predictions. The field observations represented yields measured across three soils, 2 years, and three harvest managements in New York (n = 60). The best model included factors for production potential, heat summation, soil water holding capacity, and cutting management (r2= 0.758). Relatively simple soil water budgets contributed little to model accuracy beyond what was accounted for by a general ranking of soils by water holding capacity in the humid environments. Representing the four factors of production potential, heat summation, soil water holding capacity, and management in a simulation model appears to provide a simple basis for formulating base‐line models.