Estimating surface evapotranspiration (ET) is extremely important for managing water resources in arid region. Though a number of surface energy balance (SEB) models have been proposed for estimating ET from remote sensed data, there is still no clear guidance on which models are suitable for a specific arid region. In this study, a one-source SEB model named Mapping evapotranspiration at high Resolution with Internalized Calibration (MTRIC), and a two-source SEB model named Two-source Trapezoid Model for Evapotranspiration (TTME), were selected to estimate latent heat flux (LE) in a selected sub-region of the middle reaches of the Heihe River Basin, northwest China. Both METRIC and TTME algorithms, which were developed in MATLAB environment, were applied to retrieve LE from Landsat-8 and Landsat-7 data during the crop growth stage from April to September, 2013. The derived LE with METRIC and TTME were evaluated by observed LE from ECS. The experimental results indicate: 1) The spatial distributions of LE from METRIC and TTME models are similar, with the relative deviation values within 7%. It supposes that the two models are suitable for retrieving LE in this study area. 2) LE retrieval precisions of both METRIC and TTME models are affected by land use/cover types. In desert area, though both METRIC and TTME are suitable for retrieving LE, while METRIC obtained slightly higher accurate values than TTME; in farmland/wetland area, TTME achieved obviously higher LE precision than METRIC.