The role of unforced multidecadal variability in globally averaged surface temperature evolution is investigated. A simple box climate model is employed to help quantify the relative magnitudes of forced and unforced variability over the past millennium. The apparent extent of unforced variability deduced from the historical record is contrasted to that found in retrospective as well as control runs of both CMIP 3 and CMIP 5 coupled global climate models. The spatial patterns of this variability are also investigated so that its physical origin may be better understood. The box model, along with supplementary statistical techniques are then used to create retrospective as well as future projections of surface temperature evolution with revised confidence intervals. These confidence intervals are compared to those produced from ensembles of coupled global climate models. Implications of these results on 21st century temperature projections both on the global as well as the regional level are investigated.