Proper simulation of soil temperature and permafrost at high latitudes in land surface models requires proper simulation of the processes that control snowpack development. The Simple Biosphere/Carnegie-Ames-Stanford Approach (SiBCASA) did not account for depth hoar development and wind compaction, which dominate snow processes at high latitudes. Consequently, SiBCASA had difficulty properly simulating seasonal soil freeze/thaw and permafrost. We improved simulated soil temperatures at high latitudes by (1) incorporating a snow classification scheme that includes depth hoar development and wind compaction, (2) including the effects of organic matter on soil physical properties, and (3) increasing the soil column depth. We ran test simulations at eddy covariance flux tower sites using the North American Regional Reanalysis (NARR) as input meteorology. The NARR captured the observed variability in air temperature, but tended to overestimate precipitation. These changes produced modest improvements in simulated soil temperature at the midlatitude sites because the original snow model already included the weight compaction, thermal aging, and melting processes that dominate snowpack evolution at these locations. We saw significant improvement in simulated soil temperatures and active layer depth at the high-latitude tundra and boreal forest sites. Adding snow classifications had the biggest effect on simulated soil temperatures at the tundra site while the organic soil properties had the biggest effect at the boreal forest site. Implementing snow classes, a deeper soil column, or organic soil properties separately was not sufficient to produce realistic soil temperatures and freeze/thaw processes at high latitudes. Only the combined effects of simultaneously implementing all three changes significantly improved the simulated soil temperatures and active layer depth at the tundra and boreal sites.