Quantifying ecosystem water use efficiency (WUE), its spatial variation along an aridity gradient, and its control in response to drought are crucial to understanding regional eco-physiological processes of heterogeneous landscapes. This study examined the magnitude, large-scale spatial patterns in WUE, and underlying drivers by examining published data from 31 eddy-covariance (EC) towers and remote sensing based assessments of normalized difference vegetation index (NDVI) along a southwest-to-northeast aridity gradient in China. Average growing-season WUE at the 31 sites was 1.67 +/- 0.98 g C kg(-1) H2O. Water use efficiencies among vegetation types were significantly higher for cropland and forest, and lower for shrubland and grassland ecosystems (all p-values < 0.01). Together, meteoric water and elevation explained 73% of the cross-gradient variation in WUE. Water use efficiency was shown to increase in direct association with increasing precipitation (PPT) and indirectly through PPT's effect on NDVI. Among ecosystem types, WUE for cropland was most sensitive to PPT and NDVI, whereas shrubland WUE was the least sensitive to these factors. Water use efficiency in the semi-arid zone was more sensitive to the PPT and NDVI, unlike the other climatic zones. Although growing-season PPT was important to WUE for the drier parts of China, it was less important for the dry sub-humid, moderately wetter parts of the country. The present findings showed WUE in dryland ecosystems in China to be firmly controlled by PPT. This suggests that future climate change may have differential outcomes on ecosystem carbon and water cycling along an aridity gradient, leading to the observed cross-gradient differences in ecosystem WUE's.