Gases in ice cores are invaluable archives of past environmental changes (e.g., the past atmosphere). However, gas fractionation processes after bubble closure in the firn are poorly understood, although increasing evidence indicates preferential leakages of smaller molecules (e.g., neon, oxygen, and argon) from the closed bubbles through the ice matrix. These fractionation processes are believed to be responsible for the observed millennial δO2/N2 variations in ice cores, linking ice core chronologies with orbital parameters. In this study, we investigated high-resolution δAr/N2 of the GISP2 (Greenland Ice Sheet Project 2), NGRIP (North Greenland Ice Core Project), and Dome Fuji ice cores for the past few thousand years. We find that δAr/N2 at multidecadal resolution on the "gas-age scale" in the GISP2 ice core has a significant negative correlation with accumulation rate and a positive correlation with air contents over the past 6000 years, indicating that changes in overloading pressure induced δAr/N2 fractionation in the firn. Furthermore, the GISP2 temperature and accumulation rate for the last 4000 years have nearly equal effects on δAr/N2 with sensitivities of 0.72 ± 0.1 ‰ °C−1 and −0.58 ± 0.09 ‰ (0.01 m ice year−1)−1, respectively. To understand the fractionation processes, we applied a permeation model for two different processes of bubble pressure build-up in the firn, "pressure sensitive process" (e.g., microbubbles: 0.3–3 % of air contents) with a greater sensitivity to overloading pressures and "normal bubble process". The model indicates that δAr/N2 in the bubbles under the pressure sensitive process are negatively correlated with the accumulation rate due to changes in overloading pressure. On the other hand, the normal bubbles experience only limited depletion (〈 0.5 ‰) in the firn. Colder temperatures in the firn induce more depletion in δAr/N2 through thicker firn. The pressure sensitive bubbles are so depleted in δAr/N2 at the bubble close-off depth that they dominate the total δAr/N2 changes in spite of their smaller air contents. The model also indicates that δAr/N2 of ice cores should have experienced several per mil of depletion during the storage 14–18 years after coring. Further understanding of the δAr/N2 fractionation processes in the firn, combined with nitrogen and argon isotope data, may lead to a new proxy for the past temperature and accumulation rate.