A search for the effects of anisotropy on seismic body-waves predicted by theory is described. Preliminary studies were based on long-period data from the WWSSN, HGLP and SRO networks. These showed that data from the WWSSN network are unsuitable for anisotropy studies because of features in the geometry of the recording system which lead to misalignment of the digitizer relative to the galvanometer-swing (which it is not always possible to correct) and the fact that the horizontal components are not always well matched. Digital data from the HGLP (recorded after 1976) and SRO networks are more suitable for anisotropy studies but eventually it was found that the anisotropic differences are too small to be resolved by long-period instruments. Analysis of short-period teleseismic shear-waves observed at LRSM stations located in United States and southern Canada has revealed shear-wave splitting diagnostic of anisotropy somewhere along the path. The shear-wave splitting is often seen as two separate shear-wave arrivals on the rotated horizontal components. All cases of shear-wave splitting are indicated by an abrupt change in the direction of particle-motion in the horizontal plane. A selection of seismograms and associated particlemotion diagrams is presented in order to illustrate shear-wave splitting. The polarizations of the first arrival shear-waves and the delays between the shear-wave arrivals were measured and are presented in the form of stereograms. The maximum shear-wave delay observed is 2.75 seconds and on the basis of this, we calculate the thickness of the anisotropic layer to be 248 kms for a model with 4.5% differential shearwave velocity anisotropy. For a model with much higher differential shear-wave velocity anisotropy (8.4%), the thickness of the layer is only 136 kms. Our results do not allow us to constrain the depth to the top of the anisotropic layer, although on the basis of other studies we believe the anisotropic layer to be situated immediately below the Mohorovicic discontinuity. The polarizations are broadly similar to those obtained theoretically for the y- and z-cuts of olivine, transversely isotropic olivine and mixture of transversely isotropic olivine/isotropic material. On the basis of this, we tentatively identify N50°E as a direction of symmetry and note that it is approximately parallel to the absolute motion of the North-American plate. We therefore suspect a causal relationship between plate motion and the generation of anisotropy. The most likely hypothesis is that as the continental lithosphere moves across the asthenosphere, the drag on the lithosphere sets up a horizontal compression in the direction of motion of the lithosphere relative to the asthenosphere and olivine crystals align by {Okl} [100] pencil glide so that the a-axis points into the direction of plate motion while the b and c axes form girdles perpendicular to the a-axis. This would result in transverse isotropy with the axis of symmetry horizontal, an orientation which is consistent with our results. The existence of anisotropy in the upper mantle has implications for other seismological studies. In particular, focal mechanism studies which rely solely on S-wave polarizations will be erroneous and studies of travel-time residuals will need to take account of the anisotropy.