This paper assesses the validity of common tests of the consumption CAPM. It constructs a representative consumer economy calibrated to accord with annual asset pricing data. In this economy, Friend and Blume's (1975) estimation technique seriously underestimates the degree of risk aversion of the single agent, and cannot be treated as a reliable benchmark. Further, it shows that in this economy, assuming that the large sample properties of Generalized Method of Moments (GMM) estimators are true in small samples can lead one to ‘overreject’ the model. It suggests that tests proposed by Hansen and Jagannathan (1989) may avoid these difficulties.