A time-scale decomposition (TSD) approach was introduced to statistically downscale the predictand which contains distinct variablity linked with distinct large-scale predictors. It decomposed both the predictand and the predictors into distinct components through filtering and calibrated distinct predictive equations, respectively. Due to the interannual and inter-decadal variability in July-August North China rainfall, it was used as a case to be downscaled by TSD approach. Sea level pressure, 500-hPa geopotential height, 850-hPa meridional wind, and sea surface temperature were considered as predictor parameters; several well-known large-scale climate indices were also taken as potential predictors. An approach of cross-validation-based stepwise regression was used to formulate the regression equations. The downscaling model for the interannual rainfall variability was linked to the sea surface temperature over the mid-eastern tropical Pacific in June and the 850-hPa meridional wind over East China in July－August, while that for the inter-decadal rainfall variability was related to the sea level pressure over the southwestern Indian Ocean in June under the effect of sea surface temperature over the Indian Ocean－Pacific warm pool. The downscaled interannual and inter－decadal rainfall components were added together to obtain the downscaled total rainfall. The results in the independent validation period (1991－2008) showed that the TSD approach performed well to downscale July－August North China rainfall with the correlation coefficient of 0.82 and relative root-mean-square error of 14.8%. With the hindcasted predictors by general circulation models (GCMs), the downscaling model was used to hindcast July－August North China rainfall over 1991－2001. Compared to GCM-hindcasted rainfall, the downscaling model showed better performance, which improved the original bias in terms of insufficient interannual variation in GCM hindcast. The correlation coefficient between the observed and downscaled rainfall reached 0.45, much higher than 0.12 in GCM hindcast.