Modeling the impact of small-scale land surface heterogeneities on scales resolved by general circulation models (GCMs) has long been a challenging problem. We present here a global offline comparison between two approaches to account for the heterogeneities. These approaches are mosaic, which computes separate energy budgets for each surface type within a grid box, and dominant, which assumes that a grid box can be completely described by the dominant vegetation. The experiments are all conducted using the turbulence parameterization of the Goddard Earth Observing System (GEOS) GCM, coupled to the Koster-Suarez Land Surface Model. The results show a large impact in the high- and middle-latitude Northern Hemisphere climates. At high latitudes the warming of the surface after the spring snowmelt is more rapid for dominant. At midlatitudes, where the surface is potentially under moisture stress, the mosaic approach results in a drier, warmer climate. This impact is determined to a large extent by the influence of bare soil areas on the grid-scale climate. The impact of the choice of approach is less important over more homogeneous terrains, such as deserts, as can be expected in the offline framework. These results support the need for a mosaic-type approach to properly model the coupling at the land surface interface.