A statistical downscaling scheme based on time-scale decomposition is adapted for summer rainfall prediction over the Yangtze–Huai River region of east China. The predictors are selected from atmospheric circulation variables outputted from the dynamic system models attending the Development of a European Multimodel Ensemble System for Seasonal to Interannual Prediction program (DEMETER) or observational datasets. Both the predictand and the predictors are decomposed into interannual and decadal components. Two distinct statistical downscaling models are built for the separated time scales and the predicted results are combined to represent the total prediction. The efficiency of this approach was assessed through comparisons with the models’ raw hindcasts as well as that from one parallel statistical downscaling scheme without time-scale decomposition. The results display that the time-scale decomposition scheme leads to significant improvements in the spatial and temporal correlation coefficients (CCs) and the root-mean-square errors(RMSEs) as well. The multiyear averaged spatial CCs reach up to 0.49 for all the individual models and their multimodel ensemble (MME), and the temporal CCs at each station are significantly higher with the coefficients from 0.46 to 0.7. Furthermore, two cases, the years 1998 and 1999, are selected for comparison. The former is a relatively easy predictable case and nearly all models predicted successfully, whereas the latter is a difficult case and nearly all models failed. The results suggest significant improvements for both cases. Thus, the present statistical downscaling scheme with time-scale decomposition may be appropriate for operational predictions.