In order to improve the accuracy of precipitation forecasts in the Shihe Catchment of Huaihe River Basin, the data of TIGGE models including JMA, KMA, UKMO, ECMWF and CMA along with the observed data from the rain-measuring stations in the flood season from 2015 to 2017 were used to study the evaluation methods and correction methods of precipitation forecast. On the basis of hit rate, false alarm rate and miss alarm rate, a new three-rate comprehensive index of precipitation forecast was proposed for an entire evaluation. The accuracy of TIGGE precipitation forecast in the Shihe catchment within the 1-7 d prediction period was evaluated by using the root mean square error index and the three rates comprehensive index. The correction methods of precipitation forecast, including the RBF artificial neural network and the ν-SVR support vector regression based on the data of TIGGE models, were proposed and compared with the linear method BREM. The results show that JMA model has the highest accuracy of precipitation forecast, the next two are ECMWF and UKMO, and ν-SVR method is superior to BREM method and RBF method. As a conclusion, the real-time ν-SVR correction method improves the accuracy of precipitation forecast.