Mantle plumes or decompression melting of a heterogeneous mantle have been proposed to explain oceanic-island basalt (OIB) type lavas found in the western Trans-Mexican volcanic belt. We show that mantle plumes cannot account for several geologic observations, whereas decompression melting is unrealistic given the low extension rate in the volcanic arc. On the basis of new geologic and geochemical studies, we propose an alternative model that reconciles geologic data, plate tectonic history, and petrology. Since 8.5 Ma, the volcanic front has migrated trenchward ∼80 km, and OIB began to be emplaced in the rear half of the arc after a period of stalled subduction of the Rivera plate between 7.2 and 4.8 Ma. As a whole, OIB accounts for ∼5% of the total volume of volcanism. Their trace element signatures require an enriched-mantle source, akin to the deep asthenosphere, but also indicate a minor involvement of a subduction component. We propose that during the period of very low convergence at the end of the Miocene, the leading edge of the Rivera slab started to sink into the mantle and continued rolling back until it reached the present dip of 45°. The sinking of the slab may have induced small amounts of asthenosphere to flow laterally into the opening mantle wedge, from the Gulf of California rift on the west and through the subducted Rivera-Cocos plate boundary on the east. This mechanism would produce veins of relatively enriched mantle, which would be able to produce melts with an OIB composition once the flux of fluids released from the subducted slab resumed.