Using observed meteorological data collected from four sites in the former Soviet Union, a land surface scheme designed for use in climate models was integrated for 6 years. Using observed available soil moisture and freezing front data, the model simulation of freeze-thaw cycles over the 6 years for each station was validated. The model was able to simulate the general characteristics of the seasonal freeze-thaw cycles for all stations for all years. There were problems in the simulation of the spring period by Best Approximation of Surface Exchanges (BASE): the model simulated excess accumulation of liquid soil moisture in spring and failed to match the observed timing of total freeze and total thaw. Most problems could be linked to the parameterization of hydraulic conductivity, and when this was modified, the model was able to simulate the available soil moisture more accurately. Moreover, it was shown that the inclusion of a frozen soil parameterization is necessary in order to capture the broad seasonal cycle of soil moisture and that the impact of not accounting for frozen soil can extend beyond the spring thaw period. To improve the simulations further, it may be necessary to model the effects of cryoturbation (including macropores), freezing fronts, ice- lensing and other periglacial processes. Given the spatial resolution of climate models, the vertical resolution of land surface schemes (1--5 layers), and the lack of suitable observed data, even if these processes could be parameterized, the quality of the simulations are unlikely to increase enough to warrant the additional computational expense. ¿ 1998 American Geophysical Union