Shear wave splitting due to source-side mantle anisotropy is estimated from broadband teleseismic S waves using a suite of 19 subduction zone events more than 50 km depth and recorded at several RSTN stations. We assume that there are at most two anisotropic regions along the ray path, one on the source side and one on the receiver side, and that receiver-side splitting is known from analysis of SKS. It is found that splitting is not detectable below the slab in most cases. There are, however, two specific regions, one beneath Peru and one beneath Kamchatka that indicate significant source side anisotropy. Based on two nearby intermediate focus events in each region, splitting delay times are found to be 1.0--1.5s with the fast polarization direction perpendicular to the trench. Assuming 4% intrinsic anisotropy, this would correspond to path lengths of 100--170km below the slab. Both regions are characterized by abrupt changes in along-strike slab properties, a change in slab dip in the case of Peru, and a cessation of deep focus seismicity in the case of Kamchatka. We speculate tht the anisotropy is caused by an associated local disturbance in asthenospheric flow. That the shear-wave splitting parameters of S and SKS are consistent with each other in most cases, suggests that large scale anisotropy in the lower mantle is very small, probably no more than a fraction of a percent. ¿ American Geophysical Union 1992