Freezing and thawing front (FTF) depths have implications for surface and subsurface exchanges of energy and water, vegetation growth and organic matter decomposition. Long-term changes in FTF depths are an important indicator of climate change. The FTF is seldom represented explicitly in land surface schemes, but the 0¿C isotherm is used as a surrogate for the front. However, when multiple FTFs occur within a soil column or when soil temperature hovers around the freezing point in the spring, the simulated 0¿C isotherm exhibits large fluctuations though in reality, the fronts develop rather smoothly. To explicitly simulate the FTF depths, this study couples a Two-Direction Stefan Algorithm (TDSA) in the Community Land Model 3 (CLM3). Several modifications are also introduced to adapt the CLM3 to the northern region, including the addition of a peat cover to the soil column, retention of a minimum unfrozen water content in the frozen soil, and implementation of canopy heat storage. The modified scheme was tested using field data from a boreal forest site. The TDSA enables the simulated FTF to be defined properly. Sensitivity tests demonstrate that the modified scheme (addition of a peat cover, unfrozen water and canopy heat storage) greatly improves the match between the simulated fronts and the 0¿C isotherm derived from measured soil temperatures. These modifications and the coupling of the TDSA are applicable to other lands surface schemes for the simulation of ground frost in the cold regions.