Accurate quantification of evapotranspiration (ET) at the watershed-scale remains an important research challenge. ET products from model simulations and remote sensing, even after incorporating in situ ET observations from flux towers in calibration or assimilation procedures, often produce different watershed areal-averaged ET estimates. These differences in ET estimates are magnified when they are integrated over time as part of water balance calculations. To address this challenge, we present a methodology for comparing watershed-average ET within a water balance framework that makes use of Gravity Recovery and Climate Experiment (GRACE)-observed terrestrial water storage change (TWSC). The methodology is demonstrated for South Carolina for a five-year period (2003-2007) using four different ET products: ET generated using a locally calibrated VIC model, a MODIS-derived ET product, and ET generated from two models (NOAH and VIC) as part of the North American Land Data Assimilation Systems 2 (NLDAS-2) project. The results of the example application suggest that the NLDAS-NOAH ET product is most consistent with GRACE-observed TWSC for the overall study region and time period. However, for periods of decreasing TWS, when ET becomes a more significant term in the water balance, the locally calibrated VIC model showed the most agreement with GRACE-observed TWSC. Application of the methodology for other regions and time periods can provide insight into different ET products when used for watershed-scale water resources management.