Abstract: On the basis of historical observations at 77 meteorological stations and a long-term dataset of integrated land–atmosphere interaction observations on the Tibetan Plateau (TP) (2005–2016), the historical simulations of the 12 models that participated in CMIP6 (the sixth phase of the coupled model intercomparison project), and GEWEX-SRB satellite radiation products, the authors have quantitatively examined the ability of CMIP6 models to simulate the surface sensible heat flux over the TP during 1979–2014 and analyzed the possible causes of simulation biases. The results show that CMIP6 models can well reproduce the annual cycle and seasonal spatial patterns of the sensible heat over the TP, albeit with lower amplitudes than the calculated sensible heat flux, particularly showing obvious underestimation over the strong, sensible heat regions. The area-weighted long-term spring sensible heat fluxes over the central and eastern TP simulated by the 12 models are generally lower than calculated values, with the minimum sensible heat amplitude in the MIROC6 simulation, which has approximately 1/3 of the climatological amplitude in calculated sensible heat flux. Furthermore, the authors find that the wind speed at 10-m height and the springtime land–air temperature difference simulated by models are larger and smaller than the calculated values, respectively, implying that the lower sensible heat amplitudes simulated by CMIP6 multi-models are mainly due to the cold biases of the land–air temperature difference. Spatially, the cold biases of the land–air temperature difference widely exist over the central and eastern TP, in which, more specifically, the surface and air temperatures show the cold biases but with colder biases in surface temperature. The cold biases mechanistically are likely related to the simulated stronger precipitation over the TP in CMIP6 models.