The Eocene India-Eurasia collision is a first order tectonic event whose nature and chronology remains controversial. We test two end-member collision scenarios using coupled global plate motion-subduction models. The first, conventional model, invokes a continental collision soon after ∼60 Ma between a maximum extent Greater India and an Andean-style Eurasian margin. The alternative scenario involves a collision between a minimum extent Greater India and a NeoTethyan back-arc at ∼60 Ma that is subsequently subducted along southern Lhasa at an Andean-style margin, culminating with continent-continent contact at ∼40 Ma. Our numerical models suggest the conventional scenario does not adequately reproduce mantle structure related to Tethyan convergence. The alternative scenario better reproduces the discrete slab volumes and their lateral and vertical distribution in the mantle, and is also supported by the distribution of ophiolites indicative of Tethyan intraoceanic subduction, magmatic gaps along southern Lhasa and a two-stage slowdown of India. Our models show a strong component of southward mantle return flow for the Tethyan region, suggesting that the common assumption of near-vertical slab sinking is an oversimplification with significant consequences for interpretations of seismic tomography in the context of subduction reference frames.