Recent research by Gu and Wu (2003) and Basu and Markov (2004) suggests that the well-known optimism bias in analysts’ earnings forecasts is attributable to analysts minimizing symmetric, linear loss functions when the distribution of forecast errors is skewed. An alternative explanation for forecast bias is that analysts have asymmetric loss functions. We test this alternative explanation. Theory predicts that if loss functions are asymmetric then forecast error bias depends on forecast error variance, but not necessarily on forecast error skewness. Our results confirm that the ex ante forecast error variance is a significant determinant of forecast error and that, after controlling for variance, the sign of the coefficient on forecast error skewness is opposite to that found in prior research. Our results are consistent with financial analysts having asymmetric loss functions. Further analysis reveals that forecast bias varies systematically across style portfolios formed on book-to-price and market capitalization. These firm characteristics capture systematic variation in forecast error variance and skewness. Within style portfolios, forecast error variance continues to play a dominant role in explaining forecast error.