This study examines the dynamics and robustness of large-scale evapotranspiration products in water-limited environments. Four types of ET products are tested against rainfall in two large semi-arid to arid Australian basins from 2003 to 2010: two energy balance ET methods which are forced by optical satellite retrievals from MODIS; a newly developed land surface model (AWRA); and one approach based on observations from the Gravity Recovery and Climate Experiment (GRACE) and rainfall data. The two basins are quasi (Murray-Darling Basin: 1.06 million km2) and completely (Lake Eyre Basin: 1.14 million km2) endorheic. During the study period, two extreme climatic events—the Millennium drought and the strongest La Niña event—were recorded in the basins and are used in our assessment. The two remotely-sensed ET products constrained by the energy balance tended to overestimate ET flux over water-stressed regions. They had low sensitivity to climatic extremes and poor capability to close the water balance. However, these two remotely-sensed and energy balance products demonstrated their superiority in capturing spatial features including over small-scale and complicated landscapes. AWRA and GRACE formulated in the water balance framework were more sensitive to rainfall variability and yielded more realistic ET estimates during climate extremes. GRACE demonstrated its ability to account for seasonal and inter-annual change in water storage for ET evaluation.