The Eocene-Oligocene Transition (EOT), approximately 34 Ma ago, is an interval of great interest in Earth’s climate history, due to the inception of the Antarctic ice sheet, major global cooling, and the likely inception of North Atlantic Deep Water. In line with geological evidence that the Arctic became isolated from the Atlantic at the EOT, and that the Arctic was very fresh at the time, we propose here that the onset of Atlantic deep water formation was triggered by the closing of the Atlantic gateways to the Arctic. We test this hypothesis in a new paleoclimate model, based on the GFDL CM2.1 climate model. This model is adapted to late Eocene (38 Ma) paleogeography, with a higher resolution than typically used in paleoclimate studies, i.e. the ocean and atmosphere resolutions are 1° x 1.5° and 3° x 3.75° respectively. This represents a significant step forward in resolving the ocean geography, gateways and circulation in a coupled climate model of this period. In the control experiment, we reproduce the observed low Eocene salinities (~20 psu) in the Arctic and freshwater transport from the Arctic to the Atlantic inhibits deep water formation there. Instead, bipolar sinking occurs in the North Pacific and Southern Ocean. The closing of the Arctic-Atlantic gateway causes a major salinification of the North Atlantic, triggering sinking in the North Atlantic and a corresponding shutdown of North Pacific sinking. This mechanism may explain recent proxy evidence that surface waters of the Labrador Sea became saltier and denser around the EOT. We compare the effect of the Arctic gateway closure to that of lowering CO2 and found that, while the North Atlantic salinity increases by 1-2 psu when CO2 is halved, these changes are not large enough to alter the large-scale overturning.