Atmospheric general circulation models (GCMs) forced with observed sea surface temperature (SST) reproduce some aspects of observed Sahel rainfall variability, particularly decadal variability. Here a filter based on signal-to-noise (S/N) EOFs is applied to seven GCM simulations of Sahel precipitation to extract SST-forced variability. Using filter coefficients based on GCM estimates of internal variability has limited, though positive, impact on simulation skill. Additional removal of empirically identified model error improves the representation of both decadal and interannual variability. The model error shows some coherence across the seven GCMs and correlates with local Atlantic SST. We hypothesize that the model error is related to the representation of ocean-atmosphere interactions in the SST-forced GCM simulations.