The increase in the average surface air temperature anomaly (SATA) over China is higher than the global average. However, the accurate simulation and attribution of regional SATA evolution remain challenging for current global climate models. This study simulates historical SATA variations over China using the coupled (FGOALS-g3) and uncoupled (atmospheric component, GAMIL3) models and examines their possible causes. Results show that both models reproduce the historical SATA variation with higher correlation coefficients (0.735 and 0.782) than many global climate models (0.25–0.56), although they overestimate or underestimate the changes of SATA to some extent in different periods. The results show that the cooling trend during 1941–1970 is well simulated with the coupled model while poorly presented with the uncoupled model; the coupled simulations particularly produce stronger long-term trends than the uncoupled ones during 1870–2014 considering full interaction among the atmosphere, ocean, and sea ice. In contrast, the uncoupled simulations reproduce better decadal and multi-decadal SATA variations owing to the constraints of the observed sea surface temperature (SST), such as the Atlantic multidecadal oscillation, and sea ice cover. Using Detection and Attribution Model Intercomparison Project (DAMIP) experiments, we found that the warming in the early 20th century and the recent 50 years is mainly driven by natural forcings and greenhouse gases (GHGs), whereas the cooling during 1941–1970 is caused by natural factors and anthropogenic aerosols. The cooling effects of anthropogenic aerosols are mainly attributed from the indirect SST-mediated responses through the atmosphere-ocean interactions in the coupled model.