As part of a larger study on spatial variability of land surface processes, the authors explore the sensitivity of land surface modules for climate models to the method of simulating the unsaturated subsurface flows. By examining the behavior of a number of different subsurface modules, it is shown that the surface fluxes, and consequently the water balance throughout the year, vary widely for different simulations of subsurface conditions. Typical results are presented for a specified climates and soil types. In order to reduce the complexity and computation time for the subsequent sensitivity studies, it is shown that a linearized module displays the range of behavior expected in practice. For given forcing functions of precipitation and potential evaporation, varying the depth of the modelled soil layer and changing the lower boundary conditions greatly influence the annual values of the components of the water balance. Monte Carlo simulations are used to demonstrate that spatial variation in soil properties produces large variation in runoff and compensating variations in deep drainage with a much smaller variation in evaporation. Finally, it is shown that for a given coefficient of variation in soil scaling properties, the effect on the effective large-scale sorptivity is insensitive to the type of statistical distribution used to describe the variation. ¿ 1997 American Geophysical Union