We develop a model of arc and behind-arc volcanism, as well as constraining the evolution of the lithospheric mantle structure beneath the Bolivian Andes during the last 25 Ma, using new geochronological and geochemical data on behind-arc mafic volcanic rocks, together with a regional isotopic study of geothermal helium emissions, and a comparison with seismic tomographic images of the same region. Helium isotopes measured in natural gas emissions in geothermal and mineral-water springs suggest recent mantle melting in a wide zone extending at c. 20°S from a few tens of kilometres west of the volcanic arc to over 300 km behind the arc, where the subducted slab is at depths increasing from c. 100 to c. 250 km. During the Neogene, there has been behind-arc mafic–felsic magmatism right across the high Andes between 17° and 22°S, in the Altiplano and western margin of the Eastern Cordillera. Modelling of Plio-Pleistocene basalts and basaltic andesites (<5.2 Ma), using REE, suggests that the main melt source region is in spinel and spinel–garnet peridotite, at depths <100 km. In detail, the best-fit REE inversions suggest two spatially distinct melt zones: (1) a c. 15% depleted mid-ocean ridge basalt (MORB) source at depths between c. 75 km and c. 110 km for mafic volcanic rocks that crop out <75 km behind the arc; (2) a slightly enriched MORB source at depths between c. 65 km and c. 90 km for those that crop out >100 km behind the arc. REE inversions for Oligo-Miocene shoshonites and basalts (c. 25–21 Ma) suggest that they originated at significantly shallower depths (c. 45 km to c. 100 km) from a slightly enriched MORB source. Eruption of shoshonitic lavas also occurred in the Altiplano between 13 and 11 Ma, and widespread behind-arc intermediate–felsic volcanism has been more or less continuous since the early Miocene. All this suggests that since c. 25 Ma there has been a thin (<100 km) lithosphere beneath the high part of the Bolivian Andes. Long-lived mantle melting is best explained by wet adiabatic decompression in the upwelling part of the corner flow in the mantle wedge, controlled by the lithospheric thickness variations, with depleted melt advected into the melt zone beneath the arc. Widespread arc and behind-arc magmatism may have been triggered at c. 25 Ma by the opening up of a mantle wedge and inflow of hot asthenosphere, as a consequence of the steepening of the subducted slab and detachment of hydrated and weakened mantle lithosphere and possibly mafic lower crust.