This paper discusses the relative merits of ordinary least squares estimation and several simultaneous-equations methods for econometric purposes. Most of the evidence appealed to is in the form of Monte Carlo experiments, because there is as yet little small sample theory of the properties of the different estimators.