Here, we evaluate basin-scale evapotranspiration (ET) estimates for eleven major river basins in the contiguous United States against a water balance approach with Gravity Recovery and Climate Experiment (GRACE) satellite observations. The relatively precise measurements of large-scale changes in water mass from GRACE are used to estimate the storage rate term in the terrestrial water budget and consequently provide an estimate, with propagated uncertainty, of basin-aggregated ET from mass conservation. We apply GRACE-based ET to two modeling systems (NLDAS-2 and GLDAS-2.1) comprised of five land surface models and three remote sensing-based products (MOD16, PT-JPL, and FLUXCOM) for 2003 to 2014. Both the land surface model-based and remote sensing-based ET are persistently lower than GRACE-based ET in all eleven basins tested. We also find that interannual variability is greater for GRACE-ET than the model and remote sensing products, and this is attributed to precipitation variability.