The Earth’s convecting upper mantle can be viewed as comprising three main reservoirs, beneath the Pacific, Atlantic and Indian oceans. Because of the uneven global distribution and migration of ridges and subduction zones, the surface area of the Pacific reservoir is at present contracting at about 0.6 km2 x y(r-1), while the Atlantic and Indian reservoirs are growing at about 0.45 km2 x yr(-1) and 0.15 km2 x yr(-1), respectively. Garfunkel and others have argued that there must accordingly be net mantle flow from the Pacific to the Atlantic and Indian reservoirs (in order to maintain mass balance), and Alvarez further predicted that this flow should be restricted to the few parts of the Pacific rim (here termed ‘gateways’) where there are no continental roots or subduction zones that might act as barriers to shallow mantle flow. The main Pacific gateways are, according to Alvarez, the southeast Indian Ocean, the Caribbean Sea and the Drake passage. Here we report geochemical data which confirm that there has been some outflow of Pacific mantle into the Drake passage–but probably in response to regional tectonic constraints, rather than global mass-balance requirements. We also show that a mantle domain boundary, equivalent to the Australian-Antarctic discordance, must lie between the Drake passage and the east Scotia Sea.