In this study we have analyzed the seasonal variations of global mean surface air temperature (SAT) and the surface energy budgets of 17 AR4 models. For 17 general circulation models (GCMs), we have identified considerable differences in the amplitude of seasonal cycle (A) in the global mean SAT in the pre-industrial control simulations. These differences among the models have been traced, to a large degree, to difference in their simulated clear-sky downward longwave radiation flux (Rlds)and latent heat flux (LH). Major components of seasonal changes of downward LW flux and LH are closely tied to water vapor feedback. The summertime temperature anomaly leads to sharper increase in the atmospheric water vapor content and the rates of evaporation and precipitation. The increased water vapor amount drastically increases the summertime Rlds at the Earth's surface, but the amount of changes of Rlds and LH varies dramatically from one model to another. Therefore the water vapor feedback influences the seasonal changes of SAT through its roles on the seasonal variations of Rlds and LH. This implies that the simulated seasonal change of global mean SAT might contain a clue about the sensitivity of water vapor feedback and the A of in SAT thus provides some constraint on climate sensitivity since both are subject to the same climate feedback process. The inference leads us to estimate the sensitivity of climate to a doubling of CO2 to be 3.0°C with the 95% confidence interval (1.8°C, 4.3°C).