The stable isotope ratios of atmospheric CO2 (O-18/O-16 and C-13/C-12) have been monitored since 1977 to improve our understanding of the global carbon cycle, because biosphere-atmosphere exchange fluxes affect the different atomic masses in a measurable way(1). Interpreting the O-18/O-16 variability has proved difficult, however, because oxygen isotopes in CO2 are influenced by both the carbon cycle and the water cycle(2). Previous attention focused on the decreasing O-18/O-16 ratio in the 1990s, observed by the global Cooperative Air Sampling Network of the US National Oceanic and Atmospheric Administration Earth System Research Laboratory. This decrease was attributed variously to a number of processes including an increase in Northern Hemisphere soil respiration(3); a global increase in C-4 crops at the expense of C-3 forests(4); and environmental conditions, such as atmospheric turbulence(5) and solar radiation(6), that affect CO2 exchange between leaves and the atmosphere. Here we present 30 years' worth of data on O-18/O-16 in CO2 from the Scripps Institution of Oceanography global flask network and show that the interannual variability is strongly related to the El Nino/Southern Oscillation. We suggest that the redistribution of moisture and rainfall in the tropics during an El Nino increases the O-18/O-16 ratio of precipitation and plant water, and that this signal is then passed on to atmospheric CO2 by biosphere-atmosphere gas exchange. We show how the decay time of the El Nino anomaly in this data set can be useful in constraining global gross primary production. Our analysis shows a rapid recovery from El Nino events, implying a shorter cycling time of CO2 with respect to the terrestrial biosphere and oceans than previously estimated. Our analysis suggests that current estimates of global gross primary production, of 120 petagrams of carbon per year(7), may be too low, and that a best guess of 150-175 petagrams of carbon per year better reflects the observed rapid cycling of CO2. Although still tentative, such a revision would present a new benchmark by which to evaluate global biospheric carbon cycling models.