Triplicate waveform modeling is used to resolve SH (Vs) and P (Vp) wave velocity structures in the upper mantle transition zone (TZ) beneath northwestern (NW) Tibet. Focal depth move out stacking is proposed to enhance the identification of triplicate phases, and can be used to test consistency of our data. Our results show that the Vs and Vp structures are decorrelated, and that a large Vs jump occurred across the 660-km discontinuity, with a small Vs gradient above it. Conversely, the Vp model is characterized by a relatively small contrast across the discontinuity, accompanied by a high Vp gradient in the TZ. There seem no significant depth anomalies of the 660-kin discontinuity in both models. The seismic structures in TZ beneath NW Tibet are similar to recent studies beneath the central Qiangtang and western Lhasa terrains. Taking the lower TZ structures under India as references, Vs is normal but Vp appears slightly high, and thus a high ratio of Vp/Vs was indicated beneath NW Tibet. Combined results with experiment information from mineral studies, we suggest that the differential anomalies of Vp and Vs can be attributed to a chemical heterogeneity, such as increased A1 content in the lower TZ. Considering the tectonic evolution of Tibet, the chemical heterogeneity may be associated with subduction or detachment of the Tethys oceanic slab.