S waves recorded by long-period World-Wide Standardized Seismograph Network and broadband stations in North America for deep northwest Pacific subduction zone earthquakes provide evidence for anisotropy in the lowermost mantle shear velocity structure beneath the north Pacific and Alaska. Systematic delays of up to 4 s are observed between longitudinal components (SV) and transverse components (SH) of motion for core-reflected ScS waves as well as for core-grazing and diffracted S waves. The absence of significant splitting for S waves that have turning points more than a few hundred kilometers above the core-mantle boundary indicates that anisotropy is localized within the D'' region (the lowermost portion of the mantle). SV-SH differential arrival times for both ScS and Sdiff, along with path length estimates assuming a 250 km thick D'' region, indicate spatial variations in the strength of shear wave anisotropy. The strongest anisotropy (1--1.5%) is found in the eastern part of the study area, with systematic reduction in magnitude toward the west. A transverse isotropy model can explain the data, with the velocity structure for horizontally polarized waves (VSH) having a 2--3% discontinuous shear velocity increase at the top of D'' (as proposed in earlier studies of the region) and a similar structure for S wave particle motion in the direction normal to the core-mantle boundary (VSV) but with the velocity jump at the top of D'' and the velocity within D'' being reduced from that for VSH by 0.5--1.5 km/s. Large uncertainties exist for velocity gradients above and below the velocity jump, but the requirement of a reduced VSV relative to VSH in D'' is clear. Synthetic waveforms calculated by using separate isotropic structures for SH and SV match the observations well and constrain the basic anisotropic structure, because the shear waves all traverse the region with near-grazing geometries. The study area exhibits strong lateral variations in lower mantle shear velocity structure and variable thickness of the D'' layer. Topography of the D'' layer is not well resolved (because of trade-off with volumetric heterogeneity).