It is shown that Lagged correlations for and cross-correlations between observed hemispheric-mean temperature data differ markedly from those for unforced (control-run) climate model simulations. The differences can be explained adequately by assuming that the observed data contain a significant externally forced component involving both natural (solar) and anthropogenic influences and that the global climate sensitivity is in the commonly accepted range. Solar forcing alone cannot reconcile the differences in autocorrelation structure between observations and model control-run data.