On the basis of observational evidence, it has been assumed for a long time in the seismological literature that the diffracted S wave in the deep shadow of the earth's core is polarized as SH, in good agreement with models that do not contain a well pronounced low velocity zone at the base of the mantle. While this seems often to be the case, broad band horizontal records from the GEOSCOPE network show that, for some wavepaths, the diffracted SV is strong enough to be clearly observed at distances between 107¿ and 117¿. Moreover, in some cases, the diffracted SV is shifted with respect to SH by approximately a quarter period. Such observations should put additional constraints on the structure near the core-mantle boundary. Among other possibilities, lateral variations of the gradient at the base of the mantle, allowing for regions with negative gradient, and possibly azimuthal anisotropy in D', should be considered in order to explain these data. ¿ American Geophysical Union 1989