As the atmosphere, ocean, and sea ice components of global climate models are made increasingly interactive, systematic errors or biases in one component can adversely affect the other model components. The fidelity of the component interactions is especially important in the polar regions, where many atmospheric General Circulation Models (GCMs) project an amplified climatic response to increasing concentrations of greenhouse gases. In comparing the Arctic performance of five atmospheric GCMs (GFDL, GISS, NCAR, OSU, and UKMO), we illustrate key differences in the fields most relevant to sea ice/ocean forcing: surface air temperature and sea level pressure (surface wind stress). While the amplitude of the seasonal cycle of simulated air temperature is generally realistic, biases of up to 5–10°C relative to observations are apparent over much of the Arctic. The simulated sea‐level pressure pattern varies widely from model to model, and in some cases is incompatible with the observed wind‐forcing of sea ice from the Arctic Basin to the North Atlantic via Fram Strait. The implications that these differences have for transports of salinity are signi