Recently, Bretherton and Battisti (1999) have presented an interesting interpretation of ensemble experiments with atmospheric general circulation models (AGCMs) forced by observed sea surface temperature (SST) whose mean successfully simulates the decadal evolution of the observed North Atlantic Oscillation (NAO) index. Using a linear model of atmosphere/ocean interaction, they plausibly argue that this hind-cast skill, as measured by low-pass correlations between observed and simulated indices, is consistent with the ocean mixed layer merely integrating stochastic surface heat flux forcing governed by the natural variability of the atmosphere. They go on to suggest, however, that predictability associated with middle-latitude SST anomalies is limited to timescales associated with the thermal inertia of the oceanic mixed layer (perhaps a year). Here, we include ocean circulation in a simple coupled ocean-atmosphere model and also consider hypothetical limits in which the coupled system is highly predictable at low frequencies. We find that low pass correlations between observed and simulated NAO indices, obtained from ensembles of SST-forced AGCMs, are insensitive to the predictability of the system. Thus inferences about predictability of the atmosphere-ocean system cannot be made on the basis of this measure of the hindcast skill of atmosphere-only simulations. ¿ 2000 American Geophysical Union