The applicability of RF-SVR statistical downscaling model to the simulation of extreme rainfall in flood season is discussed. The structure of the proposed downscaling model is composed of two parts, which is the rainfall state classification and the regression of rainfall amount prediction. The random forest (RF) method is used for the rainfall state classification, and the support vector regression (SVR) is used for the rainfall amount prediction. The NCEP / NCAR reanalysis data from 1961 to 2000 and the rainfall observation data of 10 stations in the Luanhe River Basin are selected for model calibration, and the data from 2001 to 2012 are used for validation. The downscaling effect of the RF-SVR statistical downscaling model is compared with that of the SVR model. The results show that the daily rainfall deviation of the Luanhe River Basin simulated by the RF-SVR model is significantly reduced, and this model can improve the simulation of extreme rainfall prediction of the basin.