The authors extract the water transport produced by the National Centers for Environmental Prediction reanalysis for a 10-yr period, 1984–93, and compare its convergence into two river basins with an independent dataset, river discharge (streamflow). Analysis focuses on two basins in the United States, the Upper Mississippi and the Ohio–Tennessee Basins, where the relatively high density of routine upper-air observations might be expected to give the reanalysis its closest rendition of the actual water transport. Over periods of several years, water input by the atmosphere should match water output from these basins in streamflow. However, in both basins an imbalance between the two with biases with respect to streamflow approaching 40% is found. The accuracy attributed to river discharge measurements averaged over several years and the apparent lack of significant multiyear storage in the basins lead us to conclude that the bias is largely an inaccuracy in the atmospheric transport. Temporal variability of atmospheric input and streamflow output shows somewhat better correspondence, with statistically significant correlations occurring for both basins on interannual and several-day timescales. The overall behavior suggests that the temporal variability of water transport depicted by the reanalysis can be used to gain insight into the actual variability of atmospheric transport, at least for well-observed regions such as the United States.