Abstract Multiyear (1987–97) limited ensemble integrations using a stretched-grid GCM, previously developed and experimented with by the authors, are employed for U.S. regional climate simulations. The ensemble members (six in total) are produced at two different regional resolutions: three members with 60-km and the other three members with 10-km regional resolution. The use of these two finer and coarser regional resolution ensemble members allows one to examine the impact of resolution on the overall quality of the simulated regional fields. For the multiyear ensemble simulations, an efficient regional downscaling to realistic mesoscales has been obtained. The ensemble means of the midtroposphere prognostic variables (height and meridional wind) show an overall good resemblance to the global reanalysis, especially for summer. Low-level features like the warm season Great Plains low-level jet are well represented in the simulations. During winter the 100-km simulations develop a southward wind east of the Rockies that is present neither in the reanalyses nor in the 60-km simulations. The analysis of the annual mean precipitation and its variance reveals that the ensemble simulations reproduce many of the observed features of a high-resolution rain gauge dataset analyzed on a 0.5° × 0.5° grid. Signal-to-noise ratios are larger than 1.5 s over a major part of the United States, especially over the Midwest and also over the mountainous regions like the Rockies and the Appalachians, suggesting that the orographic forcing is contributing to a larger signal. The ratios are smaller toward the eastern and western U.S. coastlines. This result could be attributed, at least in part, to limits in the representation of the land–sea contrasts. For comparison purposes, an additional simulation has been performed using a global uniform 2° × 2.5° grid with the same number of global grid points as those of the above stretched grids. The stretched-grid GCM ensemble means show, overall, a better regional depiction of features than those of the uniform-grid GCM. The results of the study show that even using limited ensemble integrations with a state-of-the-art stretched-grid GCM is beneficial for reducing the uncertainty of the multiyear regional climate simulation, especially when using finer 60-km regional resolution.