A suite of numerical simulations with the Center for Ocean Land Atmosphere (COLA) Studies' and the Goddard Earth Observing System (GEOS) general circulation models (GCMs) has been performed in conjunction with the Dynamical Seasonal Prediction (DSP) Project. These simulations aim to quantify the impact of realistic snow conditions on skill in the GCMs. In this study, ensemble climate simulations conducted for Northern Hemisphere spring (March--June) that span the years 1982--1998 are considered. For each year of these seasonal simulations, a pair of complementary runs is performed. For one of the simulations, snow conditions are allowed to evolve interactively; for the other simulation, the snow conditions are prescribed, according to a daily, global snow depth analysis, within each of the land modules of the GCMs. For this study, the impact of snow conditions on simulated near-surface air temperature is assessed. The results indicate that the prescribed (and presumably improved) snow conditions in the GCMs play a beneficial role in skillfully capturing the observed spatial/temporal patterns of interannual variations of near-surface air temperature, at a local scale. Through consideration of the surface energy-budget and the effect of snow cover on surface albedo, the localized improvement of near-surface air temperature skill (both in the spatial correlations and in the root-mean-square error) that results from the prescribed snowfields is found to be strongly tied to when and where the interannual variabilities of snow cover and mean incident short wave radiation coincide. This impact of prescribed snow is also most considerable during the widespread ablation of the matured winter season snow cover, which typically occurs during April over the Northern Hemisphere. Overall, the impact of the prescribed snow over all land points (i.e., local and nonlocal) shows mixed results in its effect on near-surface air temperature skill. These mixed results most likely underscore the difficulty of the GCMs to consistently translate the localized skillful response into nonlocal/remote skill. In the end, physical parameterizations in GCMs should be improved before all seasonal prediction enhancements from improved snow conditions can be realized.