The NCAR Community Atmosphere Model (CAM2) coupled with a Community Land Model (CLM2) shows a significant warm bias in surface air temperature in snow covered areas in wintertime. An intercomparison of the modeled surface albedos with MODIS-derived values reveals that in the mid-latitude northern hemisphere covered by grassland and cropland, the modeled surface albedos are much lower than observations mainly due to the shallower-than observed snow depth. This underestimated snow depth is related to how snow cover fraction (SCF) is parameterized in the land module (CLM2). Most land surface models relate SCF to grid (or patch) averaged snow depth and surface roughness length. This study suggests that the current SCF formulation of Yang et al.(1997) largely improves the simualtions of snow dpeth, surface albedo and surface air temperature over non-forested regions during wintertime. However, snow ablation is delayed and a cold bias appears in spring time. These results suggest that the SCF-snow depth relationship may be different for accumulation and ablation seasons. Guided by limited watershed-scale measurements, a different SCF-snow depth formulation for ablation season was tested in CLM2. Preliminary modeling results from CAM2/CLM2 indicated that using this SCF scheme for melting period and the SCF scheme of Yang et al. (1997) for accumualtion period leads to most encouraging results, such as significant reduction of the wintertime warm bias over snow cover areas and few undesired effects on the spring time snow depth and air temperature. The caution of using these schemes will also be discussed.