Recent studies claim that useful predictability of the North Atlantic Oscillation (NAO) up to several years in advance may be possible using atmospheric models in which the sea surface temperature (SST) is specified from observations. Achieving this goal requires that such models adequately capture the observed variation in the NAO at interannual as well as interdecadal timescales. We investigate whether this is the case by comparing interannual variability in the HadAM3 atmospheric model with observations in the SOC air-sea flux dataset for 1980--1995. We find that the model NAO time variation does not correspond to that observed, thus claims of multiannual predictability need to be viewed with caution. In addition, analysis of the observations reveals that NAO related SST anomalies do not exert a significant heat flux feedback on the atmosphere at seasonal to interannual time scales. ¿ 2001 American Geophysical Union