Abstract: Changes in actual evapotranspiration （ET_a） affect the availability of water resources and the sustainability of ecosystems. Accurately estimating ET_a and understanding its trends remain a long-standing challenge. Compared to ground-based evapotranspiration （ET） and thermal remote sensing, combining GRACE （Gravity Recovery and Climate Experiment） satellite data with the water balance equation provides a robust method to estimate long-term and basin-scale ET. Combined with streamflow observations from 72 hydrological stations, and four precipitation datasets, we calculated basin-scale ET_a via the water balance method. We evaluated eight mainstream ET products—GLDAS_NOAH, GLDAS_CLSM, GLDAS_VIC, ERA5, MERRA2, GLEAM, PML_V2, and FLUXCOM—across these basins. Results showed that annual mean ET_a from 2003 to 2020 ranged between 415.1 mm and 1001.4 mm, increasing from northwest to southeast, with an average of 626.6 mm and low uncertainty （mean: 46.7 mm）. ET product performance varied regionally, with relatively higher accuracy in northern part of China. The optimal land-surface-model-based product was GLDAS_NOAH, reanalysis product was ERA5, and remote sensing meteorological product was GLEAM. Specific recommendations included MERRA2 for the Songhuajiang River, GLDAS_CLSM for the Liaohe River and Huaihe River, GLDAS_NOAH for the Haihe River, ERA5 for the Yellow River, and GLEAM for the Yangtze River, Southeast Rivers, and Pearl River. From 2003 to 2020, ET_a increased at a rate of 5.8 mm/a, with significant upward trends in the Song-Liao River, Northern Haihe River, Southeast Rivers, Western Pearl River, and Upper Yangtze River. Land surface model selection critically influenced trend outcomes.