We argue that if analysts' objective is to provide the most accurate forecast by minimizing the mean absolute forecast error, then the optimal forecast is the median instead of the mean earnings. Forecast bias is observed when the median is different from the mean in a skewed earnings distribution. Thus, part of the observed analyst forecast bias could be a result of analysts' efforts to improve forecast accuracy when the earnings distribution is skewed. We find that earnings skewness is significantly related to analyst forecast bias. We also provide evidence that the market adjusts for part of the skewness-induced bias.