Abstract: Although deterministic forecasting is not the main purpose and application of ensemble prediction system (EPS), the forecasting performance of each individual member determines the capability of the entire EPS and the ensemble mean is also an important reference index for the actual forecasting application. Therefore, using the EPS precipitation forecast data from 2013 to 2015 (from May to October every year) from the European Centre for Medium Range Weather Forecast (ECMWF) and hourly precipitation data from CMORPH (NOAA Climate Prediction Center Morphing Method) in combination with observations collected at more than thirty thousands of automatic weather stations in China, the performance of control forecast, member forecast, and ensemble mean of ECMWF ensemble prediction system on daily precipitation in the surrounding areas of Qinling Mountains are analyzed. The effective method to improve the performance of the ensemble mean of precipitation forecast is explored. Major conclusions are as follows. (1) The spatial pattern of precipitation in the surrounding areas of Qinling Mountains is well described by the ensemble mean and the control forecast. Comparatively, the control forecast can better represent the variance of precipitation. (2) Taylor analysis shows that the precipitation variance of the ensemble mean decreases monotonously with increases in the valid period of forecast, while the variance of control forecast shows less oscillations than that of the ensemble mean and the correlation coefficient improves slightly. (3) The forecast skill scores indicate that the ensemble mean yields significantly large bias (precipitation frequency) in light rain forecast, which indicates large false alarm rate for light precipitation; meanwhile, ensemble mean decreases the bias (precipitation frequency) in heavy rain forecast, suggesting that the missing rate for heavy precipitation forecast is high. As a result, TS and ETS scores of the ensemble mean tend to be lower than those of the control forecast and disturbed member prediction, which is attributed to the skewness distribution of precipitation. (4) The significant contribution of the ensemble mean lies in its ability to well predict the spatial location of possible precipitation. By limiting the threshold, adjusting the forecast bias, decreasing (increasing) the forecast frequency on light (heavy) rain, TS, and ETS scores of the ensemble mean can be improved obviously and the ensemble forecast skill would be superior to that of the member forecast and control forecast. At present, the forecast bias correction method has been successfully applied to the fine-resolution forecast system in Shaanxi.