The estimated internal variability of near-surface air temperature was compared using three widely adopted methods [pre-industrial control (piControl) simulations, polynomial fit method, and analysis of variance method , based on 37 models from the Coupled Model Intercomparison Project Phase 5 (CMIP5) and 40 large-ensemble simulations from the Community Earth System Model (CESM). The associated influences on the time of emergence (TOE) of near-surface air temperature in future climate projections were also quantified. The results showed that for multimodels from the CMIP5, the estimated internal variability was comparable based on the piControl simulations and the polynomial fit method, while variability estimated by the analysis of variance method was exaggerated in terms of the magnitude because of its inclusion of model uncertainty. Polar amplification was evident in the spatial distribution of estimated internal variability of surface temperature, with considerably larger magnitudes in the mid- to high-latitudes than the low-latitudes. The internal variability of surface temperature did not vary significantly with time or emission scenarios, except for in the tropics, estimated by the analysis of variance method. Moreover, the estimated internal variability showed high consistency among the three methods, based on large-ensemble simulations from the CESM. The different estimated internal variabilities further affected the TOE in future climate projections, mainly in the North Atlantic Labrador Sea and the Weddell and Ross Seas in the Southern Ocean where deep overturning circulations occur. Specifically, the internal variability was estimated to be less than 15% of the forced signals over China based on all three methods in the CESM large-ensemble simulations. This result was comparable to those estimated by the piControl simulations and polynomial fit method based on the CMIP5 multimodels but tended to be overestimated by the analysis of variance method.