Global climate models are quite challenged to represent the North American Monsoon, in terms of its climatology and interannual variability. To investigate the question of whether a regional atmospheric model can improve warm season forecasts in North America, a retrospective Climate Forecast System (CFS) model reforecast (1982-2000) and the corresponding NCEP-NCAR Reanalysis have been dynamically downscaled over the contiguous U.S. and Mexico, with similar parameterization options as used for high resolution numerical weather prediction and a new spectral nudging capability. The regional model dramatically improves the climatological representation of monsoon precipitation due to its more realistic representation of the diurnal cycle of convection. However, it is challenged to capture organized, propagating convection at a distance from terrain, regardless of the boundary forcing data used. Dynamical downscaling of CFS generally yields modest improvement in surface temperature and precipitation anomaly correlations in those regions where it is already positive in the global model. There are also marked differences in potential seasonal predictability from early to late summer. CFS has a relatively greater ability to represent the large-scale atmospheric circulation in early summer due to the influence of Pacific SST forcing, and therefore surface temperature and precipitation anomaly correlations are highest at this time. As the dominant modes of early warm season precipitaiton are better represented in the regional model, provided reasonal large-scale atmospheric forcing, dynamical downscaling will add value to warm season seasonal forecasts. CFS performance appears to be inconsistent in this regard.