The actual evapotranspiration in the Jinsha River Basin was reconstructed using the GRACE observations and the GLDAS-assimilated data. A downscaling method with the constraint condition of mass conservation was used to improve the spatial resolution of GRACE data of land water storage. The sub-basin-scale monthly actual evapotranspiration in the Jinsha River Basin from 2002 to 2016 were reconstructed based on the water balance method. The spatiotemporal characteristics and dominant influencing factors of actual evapotranspiration were analyzed. The results show that the reconstructed actual evapotranspiration has high quality and compares well with seven satellite evapotranspiration products. Among them, the reconstructed actual evapotranspiration has the smallest mean deviation and root mean square error with NOAH products and the highest correlation coefficient (0.82) with PLSH products. The multi-year mean actual evapotranspiration in the Jinsha River Basin was 447.30mm, with a gradual increasing trend from the northwest to the southeast. From 2002 to 2016, the actual evapotranspiration showed an insignificant increasing trend. Of all the land use types, woodlands had the largest increasing rate of actual evapotranspiration, while grasslands had the lowest increasing rate. The actual evapotranspiration in the Jinsha River Basin was mainly affected by precipitation and temperature, followed by wind speed, and was least affected by NDVI and relative humidity.