RECENT studies of volcanic eruptions have brought to light a puzzling sequence: the ejection of dense and mixed liquids followed by a larger volume of unmixed, less dense liquid1–4. Although these studies have shown that the mixing is associated with the injection of basaltic liquid into a more silicic magma chamber, a paradox remains. Basaltic magma is denser than silicic magma, and should flow along the floor of the magma chamber. This configuration impedes mixing of the two liquids, and appears incompatible with the early eruption of basaltic liquid from the top of the chamber. Previously proposed mixing processes explain neither the eruption sequence nor the eruption of dense liquid during the time of lowest eruption rate1–8. We report here fluid-mechanical experiments that suggest a solution: a thermal plume forms over the replenishment inlet, dragging basaltic liquid upward by viscous coupling, and producing mixing only in a localized area without affecting the bulk of the chamber.