This study investigates whether and how managers manipulate earnings to improve their annual earnings forecast success, measured by predictive success, forecast error, or forecast surprise. Based on a large sample of management annual earnings forecasts from 1998 to 2006, I find strong evidence that the level of abnormal cash flow from operations and the level of abnormal production costs are associated with positive forecast errors and positive forecast surprises at earnings announcement dates. However, accruals manipulation and abnormal discretionary expenses do not affect forecast success. These results are robust after controlling for the endogeneity of forecast range. In addition, managers who issue range forecasts appear to have incentives to beat the lower bound, the midpoint, as well as the upper bound of range forecasts. Finally. managers manipulate earnings not only to reduce negative forecast surprises but also to boost positive forecast surprises at earnings announcement dates although the degree of earnings management in the presence of negative forecast surprises is more intensive.