This paper describes the impact of changes in aerodynamic roughness length for snow-covered surfaces in a land-surface scheme (LSS) on simulated runoff and evapotranspiration. The study was undertaken as the LSS in question produced widely divergent results in runoff, depending on whether it was used in uncoupled one-dimensional simulations forced by observations from the PILPS2e project, or in three-dimensional simulations coupled to an atmospheric model. The LSS was applied in two versions (LSS1 and LSS2) for both uncoupled and coupled simulations, where the only difference between the two versions was in the roughness length of latent heat used over snow-covered surfaces. The results show that feedback mechanisms in temperature and humidity in the coupled simulations were able to compensate for deficiencies in parameterizations and therefore, LSS1 and LSS2 yielded similar runoff results in this case. Since such feedback mechanisms are absent in uncoupled simulations, the two LSS versions produced very different runoff results in the uncoupled case. However, the magnitude of these feedback mechanisms is small compared to normal variability in temperature and humidity and cannot, by themselves, reveal any deficiencies in a parameterization. The conclusion we obtained is that the magnitude of the aerodynamic resistance is important to correctly simulate fluxes and runoff, but feedback mechanisms in a coupled model can partly compensate for errors. (C) 2003 Elsevier Science B.V. All rights reserved.