An Arctic region climate system model has been developed to simulate coupled interactions among the atmosphere, sea ice, ocean, and land surface of the western Arctic. The atmospheric formulation is based upon the NCAR regional climate model RegCM2, and includes the NCAR Community Climate Model Version 2 radiation scheme and the Biosphere–Atmosphere Transfer Scheme. The dynamic–thermodynamic sea ice model includes the Hibler–Flato cavitating fluid formulation and the Parkinson–Washington thermodynamic scheme linked to a mixed-layer ocean. Arctic winter and summer simulations have been performed at a 63 km resolution, driven at the boundaries by analyses compiled at the European Centre for Medium-Range Weather Forecasts. While the general spatial patterns are consistent with observations, the model shows biases when the results are examined in detail. These biases appear to be consequences in part of the lack of parameterizations of ice dynamics and the ice phase in atmospheric moist processes in winter, but appear to have other causes in summer. The inclusion of sea ice dynamics has substantial impacts on the model results for winter. Locally, the fluxes of sensible and latent heat increase by over 100 W m−2 in regions where offshore winds evacuate sea ice. Averaged over the entire domain, these effects result in root-mean-square differences of sensible heat flux and temperatures of 15 W m−2 and 2°C. Other monthly simulations have addressed the model sensitivity to the subgrid-scale moisture treatment, to ice-phase physics in the explicit moisture parameterization, and to changes in the relative humidity threshold for the autoconversion of cloud water to rainwater. The results suggest that the winter simulation is most sensitive to the inclusion of ice phase physics, which results in an increase of precipitation of approximately 50% and in a cooling of several degrees over large portions of the domain. The summer simulation shows little sensitivity to the ice phase and much stronger sensitivity to the convective parameterization, as expected.