A method to identify and to correct the large-scale biases in regional climate simulations is proposed. This bias is caused by the mismatch of the regional and the driving global models' physical parameterizations. It is suggested that simulations at spatial resolution similar to the driving global model should be used to estimate the possible bias in the higher resolution simulations. This correction is shown for a climate downscaling experiment over Southeast Asia. The global model is the NCAR community climate model (CCM3), and the regional model is the regional spectral model (RSM) of NCEP. The bias is estimated by taking the difference between a RSM0 (RSM run with similar CCM3 resolution) and the CCM3 solution. Once this bias is removed, the resulting seasonal-mean at higher resolution simulation shows a large-scale pattern more similar to that of the driving CCM3, but still retains additional realistic regional details.