Abstract: The quantitative precipitation forecast (QPF) of GRAPES-Meso4.0 model at convection-permitting resolution is evaluated thoroughly in terms of precipitation accumulation, frequency, intensity, and diurnal cycle. Real-time QPF products of the GRAPES-Meso4.0 experimental system at 3 km horizontal resolution covering southeastern China (GRAPES-Meso4.0_3 km) are verified against the observations of 24 h (08:00-08:00 BT) accumulated precipitation and hourly precipitation from 1613 stations of the National Meteorological Surface Observation Network during the summer of 2015. The QPF products over the same period and region from the operational model of the same version but at 10 km resolution (GRAPES-Meso4.0_10 km) are also used to compare and diagnose the forecast biases. Results of this study show that: (1) GRAPES-Meso4.0_3 km perfectly captured the characteristics of the total amount and spatial distribution of observed daily mean precipitation and precipitation frequency. The mean general precipitation (moderate rain and below) frequency was about 3% lower than the observation, and the precipitation frequency of heavier ones (heavy rain and above) coincided with the observation, indicating that the high resolution model can significantly correct the positive forecasting deviation of GRAPES-Meso4.0_10 km in both types of precipitation. The root mean square error (RMSE) was reduced by 40%-50%. (2) The advantage of GRAPES-Meso4.0_3 km in precipitation intensity simulation was mainly manifested in the detailed description of spatial distribution characteristics of precipitation intensity as well as the frequency and distribution of short-time heavy rainfall (precipitation intensity≥10 mm/h). However, the predicted heavy rainfall (general precipitation) intensity was stronger (weaker) than observations. (3) Diurnal cycle of hourly precipitation and precipitation frequency predicted by GRAPES-Meso4.0_3 km could reflect the observed general bimodal characteristic in the study area and the close relationship between the diurnal cycles of precipitation amount and frequency. GRAPES-Meso4.0_3 km performed better than GRAPES-Meso4.0_10 km despite the weaker peak in the afternoon (16:00 BT). (4) The model resolution is increased to convection-permitting to explicitly describe cloud and precipitation process, which is a key reason for the improvement of QPF by GRAPES-Meso4.0_3 km compared to GRAPES-Meso4.0_10 km, and the difference in the initial field of models is also a factor that cannot be ignored.