Abstract. There remain substantial uncertainties in future projections of Arctic climate change. Schmidt et al. (2014) demonstrated the potential to constrain these uncertainties using a combination of paleoclimate simulations and proxy data. They found a weak correlation between sea ice changes in the mid-Holocene (MH) and in future projections, relative to the modern period. Such an “emergent constraint” provides a powerful tool to directly reduce the range of uncertainty, provided that the necessary paleoenvironmental information is available. In the current study, we examine the relevance of Arctic warming in the past to the future through process understanding, rather than seeking a statistical relation. We conducted a surface energy balance analysis on 10 atmosphere and ocean general circulation models under the MH and future RCP4.5-scenario forcing. We found that many of the dominant processes that amplify Arctic warming from late autumn to winter are common between the two periods, despite the difference in the source of the forcing (insolation vs. greenhouse gases). We also quantified the contribution of individual processes to the inter-model variance in the surface temperature changes. The controlling term varies with the season, but the results suggest that the models’ representations of the surface albedo feedback, cloud greenhouse effect, turbulent surface heat fluxes, and indirect atmospheric stratification are important contributors. Based on the results for the Arctic warming mechanism obtained from this study, we conclude that proxy records of Arctic warming during the MH contain useful information that is relevant for understanding future Arctic climate change.