Abstract: Based on the hindcast data of MRI-CGCM of Japan Meteorological Agency, the NCEP/NCAR reanalysis data and precipitation observations in eastern China during 1979-2010, the ability of MRI-CGCM to simulate the East Asian summer monsoon (EASM) is examined. The SVD downscaling method for predicting summer rainfall in eastern China is proposed. The MRI-CGCM can reasonably reproduce the climatological summer rainfall in the Asian monsoon region. However, the simulated monsoon circulation is weaker than observations and shifts southward, which leads to underestimation of the simulated rainfall. The variance of simulated precipitation is smaller than that of observations and the ratio of its external to internal variance is lower than that of 500 hPa height, which indicates that the simulation is obviously affected by the initial condition. The simulation skill for the summer rainfall anomaly pattern over the Changjiang River valley is the highest, followed by that in South China, and the skill for simulation of North China anomaly pattern is the lowest. The model ability to simulate EOF1 of the EASM is considerably higher than that for the EOF2 simulation. The MRI-CGCM can well simulate the western North Pacific anticyclone, but underestimates its intensity. The model can reflect the effects of ENSO decaying and Indian Ocean warming on rainfall anomalies, but these effects are underestimated. For the EOF2 of EASM, the MRI-CGCM can to a certain degree realistically simulate the cyclone-anticyclone structure over the western Pacific. Only those simulations at a four-month leading time can partially reflect ENSO developing effects on rainfall. Choosing the time coefficients of the 500 hPa-heights modes through cross-validation can improve the prediction skill. The average ACCs of the SVD downscaling method are 0.20, 0.23, 0.18, and 0.02 over South China, Jiangnan, Huaihe and North China, respectively, which are significantly higher than that from the hindcast.