A statistical downscaling method is being developed at the National Centers for Environmental Prediction (NCEP) and applied to the NAEFS/NCEP global ensemble forecasts. This study is an attempt to establish the relationship between coarse and fine resolution data and project coarse resolution information onto finer grid. The operational Real-Time Mesoscale Analysis (RTMA) for CONUS with 5-km resolution (NDGD grids) is chosen as the true when compared with analysis which interpolated from NCEP Global Data Assimilation System (GDAS) to 5-km resolution. The difference between the two analyses are accumulated by adopting a decay averaging algorithm and used as a Downscaling Vector in the downscaling process. Preliminary results show that this downscaling method can effectively reduce systematic forecast errors. The downscaled forecast also displays more detailed forecast information than the raw forecast. In the next few months, NCEP is planning to implement this approach to generate NAEFS uncertainty forecast for 10%, 90% and ensemble mode at 5-km resolution.