Evapotranspiration (ET) research is rare but is essential for understanding the impacts of climate change and forest management on water resource in Nepal. This study assessed eight global remote sensing-based ET products including GLASS, GLEAM, PMLv2, REA, PEW, ETmonitor, SSEBop and SEBAL, and an ecohydrolgical model, WaSSI using locally constructed 12 watershed-scale water balance datasets. Additionally, the Budyko framework was employed to constrain ET estimates. We found serious water imbalance issues in the hydrometeorological records for medium-sized basins where measured streamflow exceeded precipitation (P) and ET/P values substantively deviated from the theoretical Budyko curve. The accuracy of the remote sensing ET products varies at selected watersheds with large elevation gradients. We found considerable modeling errors with ET overestimated, especially in high elevations among these remote sensing products. Nationally, the eight remote sensing models do not agree on the temporal trends for annual ET over the past two decades. We identified potential causes to the large errors of the ET products: 1) meteorological input data used to drive the potential ET model and ET model, 2) model algorithms causing overestimates of ET at high elevations, and 3) underestimation of precipitation and overestimation of leaf area index at high elevations. We estimated mean annual ET for High Himalya, High Mountain, Middle Mountain, Low Mountain, Terai Plain as 257 mm, 544 mm, 764 mm, 844 mm and 892 mm, respectively. This study underscores the importance of ground measuring and integrated modeling of water balances in constraining remotely estimated ET in the Himalya region.