Abstract: To improve the accuracy of the model products for precipitation forecast in flood season, we used the daily precipitation data of 92 observation stations in Jiangxi between March and July from 2015 to 2016 and precipitation forecast produced by the ECMWF (24-72 h), and chose the Gamma function to simulate the probability density distributions of precipitation and to capture the bias information, to match the probability ofthe model precipitation forecast and observation, and correct the different precipitation levels of model, thus to make a corrected precipitation value over ECMWF for every precipitation levels. The regional distribution characteristics of the model corrected precipitation value and the changes of the forecast scores before and after correction were analyzed. The results showed that: for the precipitation levels larger than 25 mm, the model precipitation forecast was less than the observation; for the low precipitation levels such as light rain, the model precipitation forecast was more than the observation; the model’s systematic errors could be effectively corrected after revising. The method of forecast-observation probability matching was effective in improving the quality of precipitation forecast, especially for the heavy precipitation forecast. The shorter the forecast time was, the better the correction would be, and it would have good application effect in business.