Abstract Earthquakes are known to occur beneath southern Tibet at depths up to ∼95 km. Whether these earthquakes occur within the lower crust thickened in the Himalayan collision or in the mantle is a matter of current debate. Here we compare vertical travel paths expressed as delay times between S and P arrivals for local events to delay times of P-to-S conversions from the Moho in receiver functions. The method removes most of the uncertainty introduced in standard analysis from using velocity models for depth location and migration. We show that deep seismicity in southern Tibet is unequivocally located beneath the Moho in the mantle. Deep seismicity in continental lithosphere occurs under normally ductile conditions and has therefore garnered interest in whether its occurrence is due to particularly cold temperatures or whether other factors are causing embrittlement of ductile material. Eclogitization in the subducting Indian crust has been proposed as a cause for the deep seismicity in this area. Our observation of seismicity in the mantle, falling below rather than within the crustal layer with proposed eclogitization, requires revisiting this concept and favors other embrittlement mechanisms that operate within mantle material.