We present an analysis of the river transport model (RTM) implementation in a fully coupled Earth system model, the Community Climate System Model, version 2 (CCSM2). The results of a 350-year CCSM2 control simulation were compared to observational river flow data and previous simulations using the Community Climate Model, version 3 (CCM3). River discharge into the Arctic Ocean improved significantly over earlier CCM3 simulations. Significant interannual variability was evident in the control simulation, although at times it exceeded the observational variability. The CCSM2 river discharges to the ocean during the peak discharge month of June for the Amazon, Amur, Ob, St. Lawrence, and Ganges Rivers were found to be within 25% of observations. CCSM2 produced negative river discharges for the St. Lawrence, Parana, and Ob Rivers for some months of the year because of the methods CCSM2 used to calculate runoff resulting from lakes, wetlands, and glaciers. Precipitation and snow cover analyses explained some of the improvements of CCSM2 over CCM3 as well as lingering sources of inaccuracy in the river discharges. The lack of water impoundment in both CCSM2 and CCM3 was shown to be another source of discrepancy between model river discharge and observed streamgage data. We discuss the differences and similarities between CCSM2 river discharges and observations within the context of extant freshwater resource management, observational measurement techniques, and spatial and temporal scales.