We have made measurements of shear wave splitting in the phases SKS and SKKS at 21 broadband stations in North America, South America, Europe, Asia, and Africa. Measurements are made using a retrieval scheme that yields the azimuth of the fast polarization direction ϕ and delay time δt of the split shear wave plus uncertainties. Detectable anisotropy was found at most stations, suggesting that it is a general feature of the subcontinental mantle. Delay times range from 0.65 s to 1.70 s and average about 1 s. Somewhat surprisingly, the largest delay time is found in the 2.7 b.y.-old Western Superior Province of the Canadian Shield. The splitting observations are interpreted in terms of the strain-induced lattice preferred orientation of mantle minerals, especially olivine. We consider three hypotheses concerning the origin of the continental anisotropy: (1) strain associated with absolute plate motion, as in the oceanic upper mantle, (2) crustal stress, and (3) the past and present internal deformation of the subcontinental upper mantle by tectonic episodes. It is found that the last hypothesis is the most successful, namely that the most recent significant episode of internal deformation appears to be the best predictor of ϕ. For stable continental regions, this is interpreted as “fossil” anisotropy, whereas for presently active regions, such as Alaska, the anisotropy reflects present-day tectonic activity. In the stable portion of North America there is a good correlation between delay time and lithospheric thickness; this is consistent with the anisotropy being localized in the subcontinental lithosphere and suggests that intrinsic anisotropy is approximately constant. The acceptance of this hypothesis has several implications for subcontinental mantle deformation. First, it argues for coherent deformation of the continental lithosphere (crust and mantle) during orogenies. This implies that the anisotropic portion of the lithosphere was present since the deformational episode and rules out the addition of undeformed material to this layer by subsequent “underplating” or conductive growth of the thermal boundary layer. One of the most important issues in the study of orogenies is the need to reconcile the formation of thickened lithosphere with the paradoxically high mantle temperatures often associated with orogenic episodes. Most efforts to date have focussed on modes of deformation whereby the cold lithospheric mantle is removed (by convective instability or delamination) and replaced by warm asthenosphere. These models, however, are incompatible with the evidence for preserved coherent lithospheric deformation; rather, the deformed mantle appears to have been heated in place. We suggest that the elevated mantle temperatures may be due to the strain heating accompanying the deformation.