Abstract: To obtain a high-quality record of actual evapotranspiration (ET_A) with which to investigate water cycle changes in the context of climate change, this study developed a method for reconstruction of ET_A based on gravity satellite observations. The proposed method used GLDAS assimilated land surface water storage data to downscale GRACE equivalent water thickness to finer spatial resolution. Then, subbasin-level monthly ET_A data for the Jinshajiang River basin were reconstructed based on the water balance during 2002—2016. Results showed that: ① The reconstructed ET_A data (ET_Recon) had high quality and compared well with three satellite-retrieval products, i.e., ET_PLSH, ET_Jung, and ET_MODIS; the strongest correlation was with the ET_PLSH product (r=0.82) and the lowest root mean squared difference was with the ET_Jung data. ② Multiyear mean ET_A in the study area was 410.8 mm/a with a gradient of increase from northwest to southeast. Temporally, it showed a significant upward trend during 2002—2016. ③ Seasonally, summer ET_A was highest and it showed an upward trend over the study period; conversely, ET_A in winter had the lowest value and it showed trivial change temporally.