Nonnormative models of aggregative agricultural production are frequently based on the correlation of output variables with prices (usually lagged) and outputs lagged over time. "Structural" models of this kind relate these correlations indirectly to the behavioral and technical structure that gives rise to them. The main purpose of this paper is to show how a knowledge of structure may be used to construct a model that simulates the aggregate actions of producers. The resulting model is useful in explaining past production patterns, in predicting future changes, and in assessing effects of alternative policies. A further purpose is to provide a theoretical understanding of the many forces that determine production through time. The author develops certain hypotheses of long standing in economic theory in an explicitly empirical and institutional context. He notes the following three as among the most important: (1) The neoclassical concept that firms maximize something when they determine output; (2) the Marshallian relation of investment to past profits and quasirents; and (3) the Schumpeterian relation between investment and technological change. To these he adds the use of behavioral constraints that arise from uncertainty and lack of knowledge. Some of the observations and principles described in the paper are empirical rediscoveries of certain theoretical principles cogently set forth by Schumpeter (12).1 The method is also closely related at some points to the work of Leontief (6, 7, 8), Wood (14), Henderson (5), Georgescu-Roegen (3), Marshall (9), Walras (13), and Nerlove (10). The empirical research out of which the new approach took form owes much to staff members of the Farm Economics Division, Economic Research Service. Thanks are due the National Cotton Council for support in the preparation of this paper.