In this paper we investigate portfolio coskewness using a quadratic market model as return generating process. It is shown that portfolios of small (large) firms have negative (positive) coskewness with market. An asset pricing model including coskewness is tested through the restrictions it imposes on the return generating process. We find evidence of an additional component in portfolios expected excess returns, which is not explained by neither covariance nor coskewness with the market. However, this unexplained component is constant across portfolios in our sample, and modest in magnitude. We investigate the implications of erroneously neglecting coskewness for testing asset pricing models, with particular interest for the empirically detected explanatory power of size