A rainstorm case in the Changjiang-Huaihe River valley on 5th July 2003 is studied in the paper,and the preprocessing and error estimation of radar radial velocity are discussed.Contrast experiments for assimilating grid radial velocity data and E-VAP(extended velocity azimuth processing) retrieved data into the nonhydrostatic meso-scale model WRF(weather research and forecasting) are performed,and results show that the direct assimilation of radial velocity improved the initial wind fields with more meso-scale information and rationalized the distribution of vertical velocity,and also improved the distribution of water vapor through the effect of winds.The assimilation of the retrieved wind data was able to increase the meso-scale information in the initial wind fields,but its effect was not as good as direct assimilation.Precipitation fields analysis shows that the direct and retrieved data assimilation both improved the rainfall forecasts,but as far as assimilation effect on precipitation concerned,the direct assimilation is better than the retrieval assimilation,and the assimilation experiments with 5—6 min assimilation interval are better than those with 30 min interval.