The authors evaluate the performance of models from Coupled Model Intercomparison Project Phase 5 (CMIP5) in simulating the historical (1951–2000) modes of interannual variability in the seasonal mean Northern Hemisphere (NH) 500 hPa geopotential height during winter (December-January-February, DJF). The analysis is done by using a variance decomposition method, which is suitable for studying patterns of interannual variability arising from intraseasonal variability and slow variability (time scales of a season or longer). Overall, compared with reanalysis data, the spatial structure and variance of the leading modes in the intraseasonal component are generally well reproduced by the CMIP5 models, with few clear differences between the models. However, there are systematic discrepancies among the models in their reproduction of the leading modes in the slow component. These modes include the dominant slow patterns, which can be seen as features of the Pacific-North American pattern, the North Atlantic Oscillation/Arctic Oscillation, and the Western Pacific pattern. An overall score is calculated to quantify how well models reproduce the three leading slow modes of variability. Ten models that reproduce the slow modes of variability relatively well are identified.