The land surface water balance of the continental United States is analyzed from 1963 to 1995 using a terrestrial biosphere model (IBIS), reanalysis data from NCEP/NCAR, a hydrologic routing model (HYDRA), and numerous observational data sets. Emphasis is placed on evaluating the performance of IBIS and the reanalysis, particularly over the central United States. IBIS is forced with daily climatic inputs from NCEP; an additional simulation is performed using observed precipitation. The NCEP reanalysis is found to have excessive precipitation and evapotranspiration over the central United States (particularly in the summertime), an exaggerated seasonal cycle of runoff, and low snow depths. The net surface water balance exhibits a dry bias that is corrected by nudging soil moisture toward climatology. Unfortunately, this correction term is large and appears to have a detrimental impact on other water balance components (particularly runoff). Fields that are reasonably well simulated in the reanalysis include fall and winter precipitation over the central United States, soil moisture in Illinois, and interannual variations in runoff. Results from the IBIS simulations show generally better agreement with observations than the NCEP reanalysis but continue to have nontrivial errors in certain fields. Over the central United States, these discrepancies include high winter/spring evapotranspiration (1 mm d-1 too high), low snow depth, and weak spring runoff (30--50% too low). The errors are at least partially caused by underestimated cloud cover and early spring green-up. A spatial analysis of the U.S. water balance reveals that some of the strongest seasonal and interannual variations in precipitation, evapotranspiration, and soil moisture occur over the central United States. ¿ 2000 American Geophysical Union