Current dynamic models are not able to provide reliable seasonal forecast of regional/local rainfall. This study aims to improve the seasonal forecast of early summer rainfall at stations in South China through statistical downscaling. A statistical downscaling model was built with Canonical Correlation Analysis method using 850 hPa zonal wind and relative humidity from the ERA-Interim reanalysis data. Anomalous southwesterly that carry sufficient water vapor encounters anomalous northeasterly from the Yangtze River, resulting in a wet anomaly in whole South China. This model provided good agreement with observations in both the training and independent test periods. In independent test, the average temporal correlation coefficient (TCC) at 14 stations was 0.52, the average root-mean-square-error was 21%. Then, the statistical downscaling model was applied to the Climate Forecast System version 2 (CFSv2) outputs to produce seasonal forecasts of rainfall for 1982–2018. Statistical downscaling model improved CFSv2 forecasts of station rainfall in South China with the average TCC increasing from 0.14 to 0.31. Forecast of South China regional average rainfall was also improved with the TCC increasing from 0.11 to 0.53. Dependence of forecast skills for regional average rainfall on the ENSO event was examined. Forecast error was reduced, but not statistical significant, when it followed El Nino event in both CFSv2 and downscaling model. And when it followed EP-type El Nino, significantly reduced forecast error (at the 0.1 level) could be seen in the downscaling model and CFSv2.