C ONSTRU CTI ON O.f large econometric models began in earnest during the years of World War II, although some earlier pioneering examples are available. The past two decades, however, mark the period of relatively large and sustained expenditures of energy and resources on the construction of these models. Currently, large econometric models are daily becoming more numerous, more "sophisticated," larger, and certainly more popular. This rising tide of popularity has been temporarily checked from time to time due to the occasional gross forecasting errors generated by these models, but on the whole these "credibility gaps" have been 'short-lived, and the increasing popularity of these devices has flowed on much like GNP and "Ole Man River." Use of large-scale econometric models for forecasting is now widespread in both government and industry; In addition, these models are used by government policy makers for the calculation of policy multipliers, the evaluation of old policies, and for the ranking of prospective new ones. Despite the central importance some of these models have assumed in decision making, the process of systematic model evaluation seems, with some noticeable exceptions, to have received relatively little attention. At the very least the problem of model evaluation has lagged behind the frenetic 250 activity in the area of model construction. Important early work in model evaluation was done by Christ [6], Adelman [1], and Theil [28], among others, and more recently the studies of Zarnowitz, Boschan and Moore [36], Evans, Haitovsky, and Treyz [11], Haitovsky and Wallace [16], as well as the current NSF-NBER Seminar on the Comparison and Evaluation of U. S. Econometric Models [9], [12], and the recent (1969) NBER Conference on Research in Income and Wealth [17], have added considerably to our understanding of this area. Nevertheless it remains true that we still lack a clear and accepted analytical basis for the selection of proper criteria for model evaluation.! Indeed, except for the simplest cases (nested hypothesis) the problem of evaluation and verification of single equation models is far from solved. In this latter area, although there are many more techniques available to aid in the process of model evaluation, there seems to be, among economists, no general agreement on the meaning and purpose of model verification.The purpose of this paper is to help clarify the notion of model verification, particularly in the empirical sciences and suggest how these considerations might "shape" our approach to the problem of the verification of large econometric models. I conclude that it is not possible to consider establishing either the truth or falsity of any theory about the structure of our economic environment, or the models that represent it, and we should instead, turn our attention to the purpose of economic theories 1 This state of affairs was, in part, the motivation for the NBER-NSF seminar on Criteria for Evaluation of Econometric Models. For this seminar's statement of the "state of the a...