Once CO2 is emitted into the atmosphere, it is gradually absorbed by natural carbon sinks. Its airborne fraction changes as a function of time depending on the sequestering performance of the sinks, in particular on their characteristic times, as reproduced by carbon cycle models. It is accepted that even after a time of the order of several thousands of years the atmospheric concentration will not return to its original value prior to the CO2 injection. The fraction of the injected CO2 amount that remains in the atmosphere for many thousands of years is measured by the atmospheric retention factor (ARF). The knowledge of this parameter is of great importance to accurately project both CO2 concentration and temperature changes in the next decades and centuries. However, current ARF estimates span a broad range from ~10 to ~30 %, clustering on average around ~20 %. Here we employ two new approaches that allow us to constrain the ARF value. The first method is based on the paleoclimatic information about the coupling between CO2 and temperature changes during the last glacial-to-interglacial transition. The second approach takes advantage of the CO2 concentration changes, that are detectable in the paleoclimatic records of the last centuries, in response to comet-induced periodic cooling episodes. We find ARF estimates of 18 and 23 %, respectively, from the two approaches. The present results confirm that the ARF is likely to be around 20 %, in agreement with pre-existing estimates.