Abstract: On the basis of the MODIS LAI remote sensing products, China Meteorological Forcing Dataset and other multi-source datasets, the Shuttleworth-Wallace (S-W) model, which considers the temporal variation of leaf area index (LAI), was applied to estimate potential evapotranspiration (ET_P) during the period of 2000—2010 in the upstream and midstream areas of the Heihe River basin. To investigate the temporal and spatial variability of ET_P, sensitivity of ET_P to meteorological parameters and LAI was calculated for each land cover type. To compare ET_P and ET₀ values for runoff simulation, a conceptual hydrologic model with precipitation and ET_P or ET₀ was developed in the upstream area of the Heihe River basin. Results showed that: ① S-W model, based on high-resolution remote sensing datasets, was able to accurately estimate annual and seasonal ET_P in large river basins with complex underlying surfaces as well as in ungauged basins. The values of ET_P in summer contributed more to the annual value than those of other seasons. ET_P in areas with different land cover types varied greatly. Generally speaking, ET_P decreased with the development of vegetation. ② In the study area, ET_P was most sensitive to relative humidity. The sensitivity of ET_P to LAI was as great as that to net solar radiation. The sensitivity of ET_P to meteorological parameters and LAI became greater with the decline of vegetation. ③ Compared with hydrologic model with ET₀ calculated by P-M model, the model with ET_P better simulated the depth of runoff at Yingluoxia hydrologic station. Hence, ET_P calculated by S-W model was more suitable for the simulation of runoff in the present study.