We present compressional (P) and shear (S) wave seismic velocity models for the upper mantle beneath southeastern Mexico derived from waveform inversion of triplicated seismic phases. The seismic waveform data produced by an earthquake located near the Mexico-Guatemala border were recorded by the La Ristra passive seismic array. The La Ristra seismic array consists of 54 broadband seismometers arranged linearly from west Texas to southeastern Utah. The orientation of the La Ristra array is nearly along the great circle from the event, and the distance (18.5¿--26.5¿) of the seismic array from southern Mexico is such that the data are ideal for investigating localized seismic structure of the upper mantle. Previous tomography and receiver function studies provide a priori knowledge of receiver-side crustal and upper mantle structure from which static adjustments were made to the seismic data. The waveforms were inverted for mantle velocity from 40 to 1000 km depth using a conjugate gradient algorithm. In the inversion, we evaluated a suite of starting models with different depths of the 410 km and 660 km discontinuities and varying velocity gradients. The best fitting models have velocity increases across the 410 km discontinuity of 6.2% and 7.3% for P and S wave velocities, respectively. The velocity jump across the 660 km discontinuity was found to be 3.3% for P waves and 6.3% for S waves. The size of the upper mantle discontinuities that we find are more in agreement with a pyrolite composition than standard reference models imply. A common feature of the best fitting models is a low-velocity zone above the 410 km discontinuity that is more prominent in the shear velocity model than the compressional velocity model. This feature may be due to partial melting induced by water release from the transition zone. The overall jump in velocity at 410 km is also larger than in previously published models with a lower gradient below. In addition, the P wave data require a small discontinuity at 490 km depth that is not resolved in the S data. Finally, the S wave data require an unusually high gradient beginning at about 600 km depth extending to the 660 km discontinuity. This feature may be due to a thermal and/or mineralogic anomaly due to a flat lying slab beneath eastern Mexico.