Seismograms of some shallow Tonga earthquakes observed at Hawaii contain SV-polarized phases in the Love wave time window, most prominently on the vertical component. Given the geometry of the observations (Δ≈40--45¿), such phases may be explained either as body waves or as mode-converted surface waves. Detailed synthetic seismogram modeling of representative events reveals several instances where the body wave explanation is inadequate, even when plausible uncertainties in the source mechanism are taken into account. The observed SV-polarized phase can instead be generated through Love-Rayleigh scattering, which requires laterally varying seismic anisotropy along the Tonga-Hawaii path. Trial-and-error forward modeling with simple structures based on the transversely isotropic mid-Pacific velocity model PA5 of Gaherty et al. <1996> obtains velocity structure that yields synthetic seismograms matching the observations. This model, while non unique, suggests first-order constraints on the lateral variation in anisotropic properties, and associated mantle flow, along the Tonga-Hawaii path. By examining trade-offs in model parameters, we conclude that robust features of the model are: (1) a transition from radial to mixed radial and azimuthal anisotropy 3¿--5¿ from Hawaii; (2) the NW-SE alignment of the axis of azimuthal anisotropy; (3) higher degree of P anisotropy relative to S anisotropy; and (4) the presence of azimuthal anisotropy within upper 200--250 km of the mantle. Taken together, these features imply a disruption of mantle fabric by the processes forming Hawaii-Emperor volcanic system. A model with anisotropic gradients in both the lithospheric lid and shallow asthenosphere is the simplest extension of our starting model. However, an equivalent data fit can be obtained if the azimuthal-anisotropy gradients are restricted to line beneath the high-velocity lid of model PA5, so that mantle hot spot flow need not penetrate the lithospheric lid. ¿ 1998 American Geophysical Union