Abstract: With the development of meteorological cause,the quality of weather forecast more and more depends on the capability of NWP products.The verification of prediction becomes a kind of efficient method of assessing forecast quality and improving the forecast.In this paper,short-term precipitation forecast by the Centre Meteorological Office's forecasters and domestic and overseas NWP models in the flood season of 2004 are verified and assessed in detail with statistical methods.The mean accumulated verification results of rainfall prediction in the flood season show that TS score of light rain for both models and forecasters is higher than that in other classes.TS scores decline sharply as rainfall increases.The difference of forecast accuracy between domestic and overseas NWP models is small.The forecast by forecasters is slightly better than that of NWP models.But for the scores of important weather processes,it appears that NWP models and forecaster are well different from each other because of the various processes.TS scores of forecaster and MM5 model are better than that of others for the severe rain in Henan province.GRAPES and Japanese models are slightly higher in skill than that of others for the severe rain caused by Typhoon 0414(unnamed).The prediction of T213 model is better than that of others for the east of Sichuan torrential rain.However,for Beijing Storm rain,it is a typical failure case that neither NWP models nor the forecaster predict correctly.Meanwhile,there are some problems in our verification system,which must be improved step by step in the future.