A 100-year off-line simulation using a land surface model (LSM) was completed. The long-term terrestrial water fluxes were estimated well using a LSM driven by long-term atmospheric forcing data that were stochastically estimated from monthly mean time series of precipitation and temperature. While high correlations between predicted and observed annual runoff are obtained at many basins globally, correlations are low in dry areas and in cool-temperate zones. These deficiencies of the simulation point out which model processes should be investigated, which can lead to improved representation of land surface processes in future LSMs from a global viewpoint. Annual snow covered area in North America and northern Europe and annual summer soil moisture in Mongolia were successfully replicated by the model. The descending trend of snow covered area in North America and Europe and the increasing trend of summer soil moisture in Mongolia, that were indicated by previous studies, were also replicated by the simulation. With the benefit of 100-year simulation results for these variables, however, these trends appear to reflect natural long-term variability rather than systematic changes in hydroclimatological condition. The long-term estimation of hydrological components, such as annual and inter-annual variation of runoff, snow and soil moisture, would be useful for examining behaviors of a LSM. The methodology used in this study also has applicability for long-term water resources assessment in poorly gauged basins, where long-term atmospheric data are only available at a monthly scale, by providing predictions of extreme behavior and long-term natural variability in various hydrological components. Copyright 2005 by the American Geophysical Union.