The vertical integrated atmospheric moisture, or the precipitable water vapor (PWV) with high temporal and spatial resolution retrieved from a ground based GPS network can supply the gap of the conventional radiosoundings. In order to evaluate the improvement of the GPS/PWV data assimilating to operat ional numerical forecasts, a continuous experiment of 3dimentional variation a ssimilation (3DVar) of GPS/PWV data into MM5 model is conducted. Through the 3DVar assimilation of GPS/PWV data, the initial field of MM5 model is built at 20:00 every day, and the hourly rainfalls over the Yangtze River Delta area a re forecasted for 24 hours. Comparing with the forecasted 6 hour's accumulative rainfalls without GPS/PWV data assimilation，the improvement effects of the MM5 precipitation forecasts with GPS/PWV data assimilation are evaluated by several test methods. The results show that the MM5 precipitation forecasts with GPS/PWV data assimilation are totally improved in the most days and most stations, espe cially in the days with smaller area of precipitation. The improvement of precip itation forecast mainly attributes to the significant decrease of the percent of false detection. The areas where the precipitation forecasts are improved signi ficantly coincide in the area with sparser regular radiosounding data.