The interannual variability in the North American monsoon and its relationship to sea surface temperature (SST) anomalies is investigated via experiments conducted with the UCLA atmospheric general circulation model (AGCM). When the model is run without interannual variations in SSTs at the lower boundary the simulation of the climatological mean monsoon is quite similar to the observed. In addition, the interannual precipitation variance and wet minus dry monsoon composite differences in the precipitation and monsoon circulation are largely realistic. When interannual variations in sea surface temperatures (SSTs) are introduced the simulated interannual precipitation variance over the southwest United States monsoon region does not increase. Nor do SSTs seem to be important in selecting for wet or dry monsoons in this simulation, as there is little correspondence between observed wet and dry monsoon years and simulated wet and dry years. These results were confirmed through a twenty-member ensemble of shorter seasonal simulations forced by an SST anomaly field corresponding to that observed for a wet minus dry southwest U.S. monsoon composite. When the AGCM is coupled to a mixed-layer ocean model the pattern of SST anomalies generated in association with wet and dry monsoons is remarkably similar to that observed: there is a large area of positive SST anomalies in the subtropical eastern Pacific Ocean and weaker negative anomalies in the mid-latitude North Pacific and Gulf of Mexico. It is demonstrated that the SST anomalies in the Pacific Ocean are forced by anomalies in the net surface solar radiative flux from the atmosphere associated with variations in planetary boundary layer stratus clouds; these variations are enhanced by a positive feedback between SST and stratus cloud variations. The anomalies in the Gulf of Mexico are associated with anomalous latent heat fluxes there. It is concluded that internal atmospheric variations are capable of: 1) producing interannual variations in the North American monsoon that are comparable to those observed, and 2) thermodynamically forcing the SST anomalies in the adjacent Pacific Ocean and Gulf of Mexico that are observed to accompany these variations. The implications of these results for seasonal forecasting are rather pessimistic since variations associated with internal atmospheric processes cannot be predicted on seasonal timescales.