Abstract: The new generation numerical weather prediction system of China, named GRAPES (the Global and Regional Assimilation and Prediction System), and the three-dimensional variational assimilation (3DVAR) system are used to assimilate radar data and to simulate a torrential rain case. The experimental results show that (1) without considering the impact of vertical motion, assimilating of radial velocity can enhance the information of mesoscale weather system in initial field and improve the prediction of early precipitation in some extent; (2) assimilating retrieved vertical velocity and rain water mixing ratio, with the vapor of ascending domain set to be saturation, conduces significant improvement of precipitation; (3) combining radial velocity with reflectivity information to assimilate is able to ameliorate initial field. The simulated 6h precipitation process is similar to the observation with the spin up phenomenon eliminated almost. The meso -βscale features are notable for the convective cell at ripe stage: its north-south width is about 50 km; the maximum of updraft vertical velocity is beyond 3.0 m/s around 500 hPa; the maximum of cloud water content is about 0.5 g/kg at the level of 600 hPa; the maximum rain water content is over 5.0 g/kg around 400 hPa; and the caused precipitation even exceeds 30 mm/h. The vapor condition of initial field generated by assimilating radar data is crucial for the meso-scale precipitation to maintain, and poor vapor condition could result in precipitation dying out promptly. At last, the professional quality control of radar data is expected.