Atmospheric general circulation models (AGCMs) forced over time with observed sea surface temperatures (SSTs) are an important tool in our ability to understand climate variability. A requirement for such integrations is global SST coverage, and this need has led to the development of new datasets constructed to take better advantage of known covarying structures in the SST field. Recent studies attempting to analyze modes of variability in the global SST field have also utilized these new data sets. But how reliable are the reconstructed SSTs, to what extent do the SSTs from separate analyses differ, and what impact do the SST differences have on AGCM simulations? We will answer these questions through an analysis of three widely used global SST data sets, two of which have been used as lower boundary conditions for an ensemble of multi-decadal integrations performed with the NCAR AGCM. The results show the differences among the SST analyses and the impact on the model climate are substantial. This has important implications for climate modeling activities around the world, as well as for the reanalysis of atmospheric data being performed at the major operational centers.