The most likely cause of seismic anisotropy in the Earth's upper mantle is the lattice preferred orientation of anisotropic minerals such as olivine. Its presence reflects dynamic processes related to formation of the lithosphere as well as to present-day tectonic motions. A powerful tool for detecting and characterizing upper-mantle anisotropy is the analysis of shear-wave splitting measurements. Because of the poor vertical resolution afforded by this type of data, however, it has remained controversial whether the splitting has a lithospheric origin that is 'frozen-in' at the time of formation of the craton, or whether the anisotropy originates primarily in the asthenosphere, and is induced by shear owing to present-day absolute plate motions. In addition, predictions from surface-wave-derived models are largely incompatible with shear-wave splitting observations. Here we show that this disagreement can be resolved by simultaneously inverting surface waveforms and shear-wave splitting data. We present evidence for the presence of two layers of anisotropy with different fast-axis orientations in the cratonic part of the North American upper mantle. At asthenospheric depths (200-400 km) the fast axis is sub-parallel to the absolute plate motion, confirming the presence of shear related to current tectonic processes, whereas in the lithosphere (80-200 km), the orientation is significantly more northerly. In the western, tectonically active, part of North America, the fast-axis direction is consistent with the absolute plate motion throughout the depth range considered, in agreement with a much thinner lithosphere.