The oceans are an important sink for anthropogenically produced CO2, and on time scales longer than a century they will be the main repository for the CO2 that humans are emitting. Our knowledge of how ocean uptake varies (regionally and temporally) and the processes that control it is currently observation-limited. Traditionally, and based on sparse observations and models at coarse resolution, ocean uptake has been thought to be relatively invariant. However, in the few places where we have enough observations to define the uptake over periods of many years or decades, it has been found to change substantially at basin scales, responding to indices of climate variability. We illustrate this for three well-studied regions: the equatorial Pacific, the Indian Ocean sector of the Southern Ocean, and the North Atlantic. A lesson to take from this is that ocean uptake is sensitive to climate (regionally, but presumably also globally). This reinforces the expectation that, as global climate changes in the future owing to human influences, ocean uptake of CO2 will respond. To evaluate and give early warning of such carbon-climate feedbacks, it is important to track trends in both ocean and land sinks for CO2. Recent coordinated observational programmes have shown that, by organization of an observing network, the atmosphere-ocean flux of CO2 can, in principle, be accurately tracked at seasonal or better resolution, over at least the Northern Hemisphere oceans. This would provide a valuable constraint on both the ocean and (by difference) land vegetation sinks for atmospheric CO2.