Abstract: The CMA-MESO (China Meteorological Administration Mesoscale Model) serves as a critical operational system for precipitation forecasting. However, its eight-times-daily rolling updates present new challenges for forecast evaluation. A comprehensive understanding of mean bias characteristics, variations across different forecast lead times, and sources of error is the key to model improvement. This study systematically evaluates the performance of CMA-MESO in precipitation forecasts over East China during the summer of 2023. The key findings are as follows: (1) While the averaged multiple forecasts generally reproduce the spatial distribution patterns of observed precipitation, systematic biases exist in precipitation intensity and frequency. (2) The model significantly overestimates the precipitation amount for moderate and above rainfall events , with a 3.7 % underestimation of no-rainfall events frequency but a 2.6 % overestimation of light rain events frequency, along with elevated false alarm rates for heavy and above rainfall events. (3) The forecast performance demonstrates distinct lead-time dependence, with precipitation biases initially decreasing and increasing with forecast lead time. The maximum biases occur in 3-6 h nowcasting, while minimum biases appear in 21-24 h forecasts. (4) Physical analysis reveals that the nowcasting overestimation stems from excessive hydrometeor introduction through cloud analysis assimilation, whereas the overestimation of forecast after 24 h results from intensified low-level wind speeds that enhance moisture transport to the target region.