Using a hierarchy of climate models together with observations from gridded analyses, I examine the atmosphere-only and coupled ocean-atmosphere variability in the general circulation for the region south of 40°S. The variability in mean sea level pressure (MSLP) is well simulated by the coupled models. A complication is that the difference between the two analyses used for verification is comparable to the analysis-model differences. An increase in variability is seen within the hierarchy of model runs although even a model without interannual variations in sea surface temperatures (SSTs) captures most of the observed variability. The temporal variation in MSLP in southern high latitudes has a white spectrum consistent with ``random'' forcing by weather events and a decoupling from oceanic ``integration''. In contrast, the spatial pattern of MSLP variability shows large-scale structure that is consistent between observations and various models, even without interannual variation in SSTs. This shows that the models are sufficiently skillful to reproduce the pattern of observed variability and suggests that the pattern of variability is a characteristic of the land-sea distribution and topography.