Past isotopic composition of atmospheric oxygen (δ 18O atm) can be inferred from the analysis of air bubbles trapped in ice caps. The longest record covers the last 420 ka (thousand of years) at the Vostok site in East Antarctica. It shows a strong modulation by the precession and striking similarities, but also noticeable differences, with the deep-sea core oxygen 18 record from which changes in the oxygen content of sea-water (δ 18O sw) and in sea-level can be derived. Indeed, δ 18O atm is driven by complex fractionation processes occuring during respiration and photosynthesis. Both δ 18O atm and its difference with respect to δ 18O sw (the Dole effect) are influenced by factors such as the ratio of oceanic and terrestrial productivities which may have significantly changed between different climates. Also, the response time of δ 18O atm to oceanic changes should be taken in consideration but this parameter itself depends on biospheric activity. We review the various aspects of the link between the δ 18O atm and the δ 18O sw signals. We also examine the approach followed by Shackleton (Science (2000)) for deriving sea-level change from the δ 18O atm Vostok record, assuming that the phase between this record and insolation changes is constant and that the Dole effect is a fraction of the precessional component of the δ 18O atm signal. Glaciological constraints on the Vostok chronology and the complexity of the Dole effect show that those two assumptions are quite probably too simplistic.