Abstract This paper characterizes and models regional-to-global–scale variability of the principal components of terrestrial water storage (TWS), which includes groundwater, soil moisture, and snow. We use the National Center for Atmospheric Research Community Land Model (CLM) that includes our recently refined parameterizations of topography-based runoff production, frozen soil hydrology, and an unconfined aquifer, to model large-scale TWS dynamics. The model is driven by global three-year (2002–2004) 3-hourly, 1 degree x 1 degree, meteorological forcing data developed by the Global Land Data Assimilation System. We use observed changes in TWS in the same period inferred from the NASA's Gravity Recovery and Climate Experiment (GRACE) satellites to evaluate the model simulations for three selected large river basins. TWS anomalies as measured by GRACE can be captured remarkably well by CLM. Groundwater dynamics in tropical regions plays a role that is different than in cold regions. In both the Niger and Amazon basins, where snow does not exist, soil moisture alone only accounts for 20%–50% of TWS anomalies, while the remaining variability is due to changes in groundwater storage. In the Ob basin, where the presence of snow is important, soil moisture and snow mass anomalies exhibit a larger amplitude than TWS anomalies, suggesting groundwater storage varies in a phase opposite to soil moisture and snow mass. In all three river basins, runoff simulations agree well with observations, and groundwater discharge dominates total runoff.