The main focus of this paper is to explore the potential for improving econometric specification in modeling hedge fund returns. Specifically, we examine the effects of (1) correcting for selectivity bias due to sample attrition; (2) allowing for nonlinearity; and (3) controlling for fund-specific unobserved heterogeneity. Diagnostic tests confirm the importance of these complications. Using data covering 1996–2008, we show that when selectivity, nonlinearity, and fund heterogeneity are taken into account, we obtain more robust estimates of the effect of key variables on hedge fund returns and managerial efficiency, and the explanatory power of the model is significantly improved. Copyright © 2013 John Wiley & Sons, Ltd.