In this paper we argue that in realistically calibrated two period general equilibrium models with incomplete markets CAPM-pricing provides a good benchmark for equilibrium prices even when agents are not mean-variance optimizers and returns are not normally distributed. We numerically approximate equilibria for a variety of different specifications for preferences, endowments and dividends and compare the equilibrium prices and portfolio-holdings to the predictions of the CAPM. While the CAPM does not hold exactly for the chosen specification, it turns out that pricing errors are extremely small. Furthermore, two-fund separation holds approximately.