P waves with travel time residuals between 0 s and -12 s are observed at regional stations in Samoa (AFI) and Raoul Island (RAO) for 39 earthquakes in Tonga with focal depths between 70 km and 300 km. These anomalously large residuals apparently are produced because seismic phases travel along the strike of the Tonga Benioff zone within the high-velocity subducted lithosphere for up to 1200 km before arriving at AFI and RAO. To eliminate erroneous residuals caused by poor event locations, we selected 11 stations and reread the available P times at these stations for the 39 events. These arrivals and (pP-P) intervals were used to relocate the events using variant of the joint hypocenter determination method. Then the pattern of residuals at AFI and RAO (not used in the relocation) was compared to the pattern of residuals expected for various models of the subducted lithosphere, as determined by ray tracing. The observed pattern of residuals at AFI is consistent with the ray-traced models if some of the first arrivals are produced by rays traveling directly along the strike of the subducting lithosphere, and if others are produced by rays which reflected once off the upper surface of the subducting lithosphere before arriving at AFI. The observed residuals can be explained by a model where the P velocity in the subducted lithosphere is 8% higher than the velocity in the Herrin model. The residuals are fit even better by a layered slab model in which the seismic velocity is about 6% higher than the Herrin velocity at the upper surface of the subducted lithosphere and about 9% higher at the bottom of the slab. These velocity contrasts could be produced if the temperature in the slab was 700 ¿C cooler than the surrounding mantle, and if there were no partially melted material within the slab. The existence of these anomalously large residuals suggest that the high velocity region in the upper 300 km of the mantle beneath Tonga must be fairly continuous over distances of 1000 km and more.