The origin of the Yellowstone and Snake River Plain volcanism has been strongly debated. The mantle plume model successfully explains the age-progressive volcanic track, but a deep plume structure has been absent in seismic imaging. Here I apply diffractional tomography to receiver functions recorded at USArray stations to map high-resolution topography of mantle transition-zone discontinuities. The images reveal a trail of anomalies that closely follow the surface hotspot track and correlate well with a seismic wave-speed gap in the subducting Farallon slab. This observation contradicts the plume model, which requires anomalies in the mid mantle to be confined in a narrow region directly beneath the present-day Yellowstone caldera. I propose an alternative interpretation of the Yellowstone volcanism. About 16 million years ago, a section of young slab that had broken off from a subducted spreading centre in the mantle first penetrated the 660 km discontinuity beneath Oregon and Idaho, and pulled down older stagnant slab. Slab tearing occurred along pre-existing fracture zones and propagated northeastward. This reversed-polarity subduction generated passive upwellings from the lower mantle, which ascended through a water-rich mantle transition zone to produce melting and age-progressive volcanism.