In this paper two continuous 3½-year-long climate simulations over the continental United States are discussed, one for present-day conditions and one for conditions under double carbon dioxide concentration, conducted with a limited area model (LAM) nested in a general circulation model (GCM). The models used are a version of the NCAR Community Climate Model (CCM) at rhomboidal 15 spectral resolution and the climate version of the NCAR/Penn State mesoscale model (MM4) at 60-km gridpoint spacing. For present-day conditions the model temperatures are within 1°–2°C of observations except over the Great Lakes region, where temperature is overpredicted by 4°–6°C. The CCM overpredicts precipitation throughout the continental United States (overall by about 60%) and especially over the West (by up to 300%). The nested MM4 overpredicts precipitation over the West but underpredicts it over the eastern United States. In addition, it produces a large amount of topographically and lake-induced sub-GCM grid-scale detail that, especially during the cold season, compares well with available high-resolution climate data. Overall, based on a number of objective measures of climate simulation skill, the nested MM4 reproduces observed spatial and seasonal precipitation patterns better than the driving CCM. Doubled carbon dioxide-induced temperature change scenarios produced by the two models generally differ by less than several tenths of a degree except over the Great Lakes region where, because of the presence of the lakes in the nested model, the two model scenarios differ by more than one degree. Conversely, precipitation change scenarios from the two model simulations can locally differ in magnitude, sign, spatial, and seasonal detail. These differences are associated with topographical features in the MM4, such as the presence of steep coastal ranges in the western United States. This work illustrates the feasibility of the use of the nested modeling technique for long-term regional climate simulation.