Abstract: The assimilation of radar data using Gridpoint Statistical Interpolation (GSI) analysis system was employed for prediction of “7·21” Beijing heavy rainfall. It is shown that the assimilation of radar data introduces reasonable modification to the initial analysis. The vapor increment associated with the assimilation of radar reflectivity, and the modified horizontal wind fi eld due to the assimilated radar radial velocity provided a better initial analysis for precipitation prediction. This may effectively improve the prediction of the spatial distribution and the intensity of heavy rain. Thus the sensitivity simulations of “7·21” Beijing heavy rainfall validate a good effect of the assimilation of both radar radial velocity and reflectivity on the heavy rainfall prediction.