The seismic structure of the transition zone beneath the northwestern Pacific ocean is studied by comparing regional distance seismic waveforms with synthetic seismograms. By modeling waveforms we incorporate the travel times of secondary waves that are not included in traveltime data used in most P wave tomography experiments. These secondary waves are highly sensitive to the velocity above the 660 km discontinuity and thus add information on the seismic structure of the transition zone independent of that used in many tomography experiments. We used just two one-dimensional models in the waveform comparison. The iasp91 (slightly modified in the shallower depths as aiasp) is used as a global average structure, and model M3.11 is used as representative of transition zone structure near a deep subducting slab. Model M3.11 is faster than iasp91 from 525 to 660 km depth but also has a 30 km deeper 660 km discontinuity. Our results show strong variations in the transition zone structure beneath the northwestern Pacific. Data that sample beneath the northwestern Japanese subduction zone and the southern Kuriles are well fit by model M3.11. The deep structure beneath the Bohai Sea, to the west of Korea, also appears to show slab-related velocity anomalies in the transition zone. In contrast, the seismic data sampling beneath the Yellow Sea to the East China Sea are better modeled using the global average model iasp91. Data that sampled beneath the Philippine Sea showed evidence for anomalously high velocity in the transition zone, but there is little apparent change in the depth of the 660 km discontinuity there. The overall results are in agreement with past work indicating fairly broad regions where subducting plate is lying flat or piling up within the transition zone, although we find that the region where this occurs, and thus the total volume of slab within the upper mantle, is considerably less than that seen in past studies. The unusual structure beneath the Philippine plate may be due to the relatively recent descent of slab beneath this region. The slab itself is cold, and thus anomalously fast, but the deeper mantle may not have cooled enough for the 660 km discontinuity to have deepened. ¿ 1998 American Geophysical Union