Four simulations of the region 30 ¿N--90 ¿N are performed using the Global Soil Wetness Project methodology and a single land surface scheme. Four methods are used to represent soil ice: an explicit representation of the thermal and hydrological effects of soil ice; two implicit methods (which only account for the hydrological effects); and finally the simplest approach where soil ice is not accounted for. Substantial impacts on total runoff, evaporation and temperature result from the choice of parameterization. The partitioning of total runoff between drainage and surface runoff is also changed. The impacts on temperature are large enough to cause problems for fingerprinting of global change while the change in the runoff generation process, and the timing of maximum runoff are large enough to concern ocean modelers. Evidence presented here and elsewhere indicates that land surface schemes should include the thermal effects of soil ice melting and freezing. However, the hydrological effects of soil ice suppressing infiltration and encouraging surface runoff may be based on observations taken at a scale inappropriate to climate model parameterization. We show that for one basin, the Mackenzie, a land surface model which ignores soil ice entirely simulates runoff better than the other methodologies tested here. We therefore hypothesize that it may be preferable to not include soil ice in the runoff formulations used in land surface models until we have more observations at an appropriate spatial scale. Testing of frozen soil moisture parameterizations in other catchments with high-quality observed runoff data should be conducted to test this hypothesis. ¿ 1999 American Geophysical Union