In this paper, four cases on the YangtzeHuaihe River Valley (Jianghuai for short) during continuous rainstorm process in July 2007 are chosen for numerical prediction experiments using the GRAPES3Dvar model with radar direct assimilation function. The results have shown that the direct assimilation of the Doppler radial velocity, as well as the basic reflectivity factor can effectively improve the numerical prediction effects of mesoscale precipitation. The further analysis of the joint and the separate assimilation of radar data assimilation on July 3 shows that: Joint assimilation can improve the initial analysis of the wind field and humidity information, stimulate model precipitation, reduce the spinup phenomenon. Hence the mesoscale precipitation’s occurrence and development as well as the strength, movements and distribution are more accurately forecasted, and the effects of mesoscale numerical forecast of heavy precipitation are significantly improved. The main contribution of radial velocity direct assimilation is in the wind field, which can increase small and mediumscale wind information such as cyclonic vortex in the initial field, improve the analysis of threedimensional wind field. The main impact of reflectivity’s direct assimilation is in the humidity field, which improves the initial humidity field parameters, and impacts significantly the precipitation forecast. 