North American Land Data Assimilation System (NLDAS) phase 2 has generated 31-year (1 January 1979 to 31 December 2009) records of water fluxes, energy fluxes and state variables from four state-of-the-art land models (NCEP/Noah, NASA/Mosaic, OHD/SAC, Princeton and Washington/VIC). In this study, these products are used to estimate long-term (1979-2008), basin-scale evapotranspiration. The suggested novel approach uses observed gauge precipitation (Po) with PRISM corrections, observed U.S. Geological Survey (USGS) streamflow (Qo), and simulated total water storage (dWm/dt) to estimate annual evapotranspiration. For a basin and individual year/month, in the real world dWo/dt =Po-Qo-Eo, while in the model world dWm/dt=Pm-Qm-Em. As dWo/dt is unknown, we assume dWo/dt can be approximately represented by dWm/dt. Therefore, (Po-Qo-Eo) will approximately equal (Pm-Em-Qm), and so Eo=Po-Pm+Em+Qm-Qo. In our NLDAS study case, Po-Pm=0 since our model uses CPC observed gauge precipitation. As a best estimation of Eo, Ee=Em+Qm-Qo. The estimated annual evapotranspiration (Ee) is compared with NLDAS-based simulated evapotranspiration (Em) and with evapotranspiration derived from application of the long-term mean water balance equation (Eb =Po -Qo), two reference evapotranspiration datasets (gridded Fluxnet evapotranspiration derived from Jung et al. (2010), and MODIS evapotranspiration) over 961 basins. The results show that Ee is closer to Eb and two reference evapotranspiration datasets than Em for all four models and their ensemble mean. A ratio between NLDAS-based evapotranspiration spread (standard deviation of ensemble mean) and estimated evapotranspiration spread is used to represent the change in evapotranspiration uncertainty. The results suggest that for annual evapotranspiration, the water budget method greatly reduced the uncertainty of modeled values over most of CONUS except for low total runoff regions, and for monthly evapotranspiration, it greatly reduced the uncertainty of modeled values in the Northeast of CONUS only; however, over other parts of the CONUS such as the West, the approach greatly increased the uncertainty of the modeled ET values. The main reason is that land surface models cannot capture the correct seasonal cycle for streamflow. The difference between Qm and Qo increased the uncertainty of the estimated evapotranspiration. Therefore, long-term accurate monthly ET estimation will depend on the improvement of land surface models in simulating the correct season cycle of streamflow. Finally, long-term estimated and NLDAS-based annual evapotranspiration is used to analyze annual evapotranspiration trends over CONUS using the Mann-Kendall test.