Biological dinitrogen fixation provides the largest input of nitrogen to the oceans, therefore exerting important control on the ocean's nitrogen inventory and primary productivity. Nitrogen-isotope data from ocean sediments suggest that the marine-nitrogen inventory has been balanced for the past 3,000 years (ref. 4). Producing a balanced marine-nitrogen budget based on direct measurements has proved difficult, however, with nitrogen loss exceeding the gain from dinitrogen fixation by approximately 200 Tg N yr−1 (refs 5, 6). Here we present data from the Atlantic Ocean and show that the most widely used method of measuring oceanic N2-fixation rates underestimates the contribution of N2-fixing microorganisms (diazotrophs) relative to a newly developed method. Using molecular techniques to quantify the abundance of specific clades of diazotrophs in parallel with rates of 15N2 incorporation into particulate organic matter, we suggest that the difference between N2-fixation rates measured with the established method and those measured with the new method can be related to the composition of the diazotrophic community. Our data show that in areas dominated by Trichodesmium, the established method underestimates N2-fixation rates by an average of 62%. We also find that the newly developed method yields N2-fixation rates more than six times higher than those from the established method when unicellular, symbiotic cyanobacteria and γ-proteobacteria dominate the diazotrophic community. On the basis of average areal rates measured over the Atlantic Ocean, we calculated basin-wide N2-fixation rates of 14 ± 1 Tg N yr−1 and 24 ±1 Tg N yr−1 for the established and new methods, respectively. If our findings can be extrapolated to other ocean basins, this suggests that the global marine N2-fixation rate derived from direct measurements may increase from 103 ± 8 Tg N yr−1 to 177 ± 8 Tg N yr−1, and that the contribution of N2 fixers other than Trichodesmium is much more significant than was previously thought.