The numerical method for estimating soil moisture transport coefficients in land surface models (LSMs) is evaluated through comparison to a weighted-average method based on successive steady states <Warrick, 1991> and a proposed new scheme <Dong and Wang, 1997>. The focus is on situations with large gradients of soil moisture. The reference cases used to evaluate the performance of the various methods are based on numerical integrations with fine spatial resolution and small time steps. A series of tests were performed to detect sources of discrepancies and to assess the impact of various numerical methods on the simulation of evaporation, transpiration, and drainage. Off-line LSM simulations driven by observed atmospheric forcings were analyzed to understand the propagation of errors between the various hydrologic processes. We suggest a simple modification to LSM that can improve the precision of simulated hydrologic fluxes without increasing vertical resolution. ¿ 2001 American Geophysical Union