Twenty years ago, measurements on ice cores showed that the concentration of carbon dioxide in the atmosphere was lower during ice ages than it is today. As yet, there is no broadly accepted explanation for this difference. Current investigations focus on the ocean's biological pump', the sequestration of carbon in the ocean interior by the rain of organic carbon out of the surface ocean, and its effect on the burial of calcium carbonate in marine sediments. Some researchers surmise that the whole-ocean reservoir of algal nutrients was larger during glacial times, strengthening the biological pump at low latitudes, where these nutrients are currently limiting. Others propose that the biological pump was more ef®cient during glacial times because of more complete utilization of nutrients at high latitudes, where much of the nutrient supply currently goes unused. We present a version of the latter hypothesis that focuses on the open ocean surrounding Antarctica, involving both the biology and physics of that region.T he past two million years have been characterized by large cyclic variations in climate and glaciation. During cold ice age' periods, large continental ice sheets cover much of the polar Northern Hemisphere. During intervening warm periods, or interglacials' , Northern Hemisphere glaciation wanes drastically. The ultimate pacing of these glacial cycles is statistically linked to cyclic changes in the orbital parameters of the Earth, with characteristic frequencies of roughly 100, 41 and 23 kyr (refs 1, 2). These orbitally driven variations in the seasonal and spatial distribution of solar radiation incident on the Earth's surface, known as the Milankovitch cycles' after their discoverer, are thought to be the fundamental drivers of glacial/ interglacial oscillations. However, the direct energy budget effects of the orbital variations are insuf®cient to drive the large amplitude of the glacial cycles that are observed, and orbital variations alone do not provide an obvious cause of the rapid climate transitions evident in palaeoclimatic and palaeoceanographic records. Positive feedbacks within the Earth's climate system must amplify orbital forcing to produce glacial cycles, but the operation of these internal feedbacks is poorly understood. One of the central goals of Earth science is to develop a mechanistic understanding of the Earth's climate feedbacks, the roles that they have played over glacial cycles, and the roles that we should expect them to play in the future.The concentration of carbon dioxide (CO 2 ) in the atmosphere has varied in step with glacial/interglacial cycles 3,4 (Fig. ). During interglacial times, such as the Holocene (roughly the past 10,000 years), the atmospheric partial pressure of CO 2 (p CO 2 ) is typically near 280 parts per million by volume (p.p.m.v.). During peak glacial times, such as the Last Glacial Maximum about 18,000 years ago, atmospheric p CO 2 is 180±200 p.p.m.v., or roughly 80±100 p.p.m.v. lower. CO 2 is a greenhouse gas, and model calculations suggest that its changes play a signi®cant role in the energetics of glacial/interglacial climate change 5 . However, we have not yet identi®ed the cause of these variations in CO 2 . How do the orbitally driven cycles in solar radiation set this particular positive feedback into motion?This question has persisted for two decades, motivating intensive research by palaeoclimatologists and palaeoceanographers. From this research, we now recognize many processes that could regulate atmospheric CO 2 on the timescales of glacial/interglacial transitions, and it may well be that the actual cause of glacial/interglacial CO 2 change is among them. However, evaluating the importance of these mechanisms with data from the recent geological and glaciological record has been a challenging and controversial task, leading as yet to no consensus on a fundamental mechanism. It is possible that many factors contribute comparably to the observed changes. However, the regularity of the CO 2 variations and the consistency of the upper and lower limits of atmospheric CO 2 through multiple 100-kyr cycles (Fig. ) are suggestive of a wellordered set of dominant mechanisms, the `holy grail' of glacial/ interglacial CO 2 research.