Slab melting has been suggested as a likely source of adakitic arc magmas (i.e., andesitic and dacitic magmas strongly depleted in Y and heavy rare earth elements). Existing numerical and petrologic models, however, restrict partial melting to very young (≤ 5 Ma) oceanic crust (typically at 60–80 km depth). Paradoxically, most of the known Pliocene-Quaternary adakite occurrences are related to subduction of 10–45 Ma lithosphere, which should not be able to melt under normal subduction-zone thermal gradients. We propose an unusual mode of subduction known as flat subduction, occurring in ∼10% of the world's convergent margins, that can produce the temperature and pressure conditions necessary for fusion of moderately old oceanic crust. Of the 10 known flat subduction regions worldwide, eight are linked to present or recent (<6 Ma) occurrences of adakitic magmas. Observations from Chile, Ecuador, and Costa Rica suggest a three-stage evolution: (1) steep subduction produces a narrow calc-alkaline arc, typically ∼300 km from the trench, above the asthenospheric wedge; (2) once flat subduction begins, the lower plate travels several hundred kilometers at nearly the same depth, thus remaining in a pressure-temperature window allowing slab melting over this broad distance; and (3) once flat subduction continues for several million years, the asthenospheric wedge disappears, and a volcanic gap results, as in modern-day central Chile or Peru. The proposed hypothesis, which reconciles thermal models with geochemical observations, has broad implications for the study of arc magmatism and for the thermal evolution of convergent margins.