Using a linear inversion method applied to regional surface wave data, we have determined seismic moment tensors for 50 earthquakes in the western United States. Regional surface waves are well suited for studying moderate-sized earthquakes (4<M<5) located outside of local seismic networks for which first-motion solutions are difficult to obtain and seismic moment estimates are problematic due to clipping of local P and S wave recordings. We compared the inversion results to P wave first-motion solutions for 17 of the 50 events, and our comparison shows much better agreement in the orientations of the T axis than for the P axis. This could be related to the sensitivity of both methods to focal depth estimates. The surface wave results yielded a very uniform and consistent data set of stress orientations. The T axis is uniformly horizontal for all areas sampled, with E-W extension across northern California, northern Nevada, and into north central Utah; ESE-WNW extension in southern Nevada; and ENE-WSW extension in Idaho, north of the Snake River Plain. The P axis shows more variability than the T axis on local as well as regional distance scales. In the Great Basin, the results indicate a N-S change in the relative magnitude of the vertical to maximum horizontal compressive stress, with most earthquakes showing normal faulting in the Battle Mountain region and strike-slip faulting in southern Nevada. The direction of maximum horizontal compressive stress is mainly N-S in northern California and western Oregon. Our results, along with the results of Zoback et al. (1981) for central California, indicate a 60¿ rotation of the maximum horizontal compressive stress between northern and central segments of the San Andreas fault system. A change in stress state along the northern segment could be related to the thermal and mechanical evolution of the Pacific-North American plate boundary. ¿American Geophysical Union 1991