Wintertime incursions of extreme Arctic air masses into middle latitudes represent an important source of meteorological variability. These socially and economically significant events cause numerous deaths every winter and result in expensive losses to the agricultural and transportation sectors. The equatorward migration of these polar air masses also constitutes an important component of the Arctic energy budget, serving as one of the sources of poleward atmospheric energy transport to high latitudes. We are investigating the properties of extreme cold-air outbreaks (CAOs) that reach middle latitudes, as simulated by various climate models. Analysis of both atmosphere-alone and coupled models reveals several interesting features about these extreme events. First, models seem capable of simulating the synoptic patterns associated with CAOs over North America (upper-air flow patterns, SLP variations, and temperature anomalies), and these patterns often differ greatly from one event to another. Second, the simulated CAO characteristics in the fixed-SST runs compare favorably with observational station data at a test site (Madison, WI) in terms of the frequency, duration, and intensity of cold air outbreaks. Third, although the simulated upper-air circulation anomalies are sensitive to SST variations associated with ENSO, there is little correlation between ENSO and CAOs over the Midwest in the models. Fourth, topographic steering of cold air surges in the western U.S. occurs in the models, but the magnitude appears to be too weak, most likely because of the models’ smoothed orography. Fifth, both observations and simulations show distinct regional variations in the skewness of wintertime surface temperature distributions that are indicative of CAOs. Sixth, greenhouse forcing seems to cause surprisingly little change in the frequency and/or intensity of CAOs in the eastern U. S. in some models. We hope that this research will eventually help identify precursor conditions that might be used to improve operational forecasts of CAOs.