The stable carbon isotopic signature of carbon dioxide (δ13CO2) measured in the air occlusions of polar ice provides important constraints on the carbon cycle in past climates. In order to exploit this information for previous glacial periods, one must use deep, clathrated ice, where the occluded air is preserved not in bubbles but in the form of air hydrates. Therefore, it must be established whether the original atmospheric δ13CO2 signature can be reconstructed from clathrated ice. We present a comparative study using coeval bubbly ice from Berkner Island and ice from the bubble-clathrate transformation zone (BCTZ) of EPICA Dome C (EDC). In the EDC samples the gas is partitioned into clathrates and remaining bubbles as shown by erroneously low and scattered CO2 concentration values, presenting a worst-case test for δ13CO2 reconstructions. Even so, the reconstructed atmospheric δ13CO2 values show only slightly larger scatter. The difference to data from coeval bubbly ice is statistically significant. However, the 0.16‰ magnitude of the offset is small for practical purposes, especially in light of uncertainty from non-uniform corrections for diffusion related fractionation that could contribute to the discrepancy. Our results are promising for palaeo-atmospheric studies of δ13CO2 using a ball mill dry extraction technique below the BCTZ of ice cores, where gas is not subject to fractionation into microfractures and between clathrate and bubble reservoirs. © 2011 Elsevier B.V.