We analyze residual spheres for teleseismic S wave arrival times from 10 intermediate and deep focus earthquakes in the Kurile subduction zone. Long-period WWSSN and CSN recordings of transverse component S and ScS arrivals provide good azimuthal coverage for each event, with arrival time estimates that compare well with short-period observations. In order to isolate the near-source structure, substantial lower mantle and receiver path effects are suppressed both by using empirical station/path statics derived from S, sS, ScS, and sScS arrivals for a total of 28 northwestern Pacific events and by constructing differential residual spheres. Ray tracing through three-dimensional aspherical shear velocity models supports the inferred large magnitude of the deep mantle and near receiver effects, although existing models do not appear to provide accurate corrections. The events are relocated in the PREM structure, and a modest spatial smoothing of the residuals is applied. The resulting residual spheres are modeled using a finite difference travel time method that accounts for three-dimensional ray path perturbations. Our starting models are derived from slab structures determined by previous P wave residual sphere analyses. We explore the compatibility of the P and S wave observations to constrain the Vs/Vp velocity ratio of the slab and to test whether the S wave arrival times place additional constraints on the slab geometry. The modeling supports the previous results favoring penetration of the Kurile slab to at least 800 km depth for most of the length of the arc, with some evidence that the northern portion of the slab broadens by as much as a factor of 3 upon entering the lower mantle. The preferred slab shear velocity models have relatively low (2--3%) shear velocity contrasts below 670 km depth, which can be accounted for by slab temperature reductions of 300¿--400¿. ¿ 1999 American Geophysical Union