A study of teleseismic P-wave travel times has been made using data from deep (450 to 650 km) earthquakes exclusively, for which the effect of near-source upper mantle heterogeneity is expected to be small and location parameters are considered reliable. About 3,300 arrival times selected from bulletins and winnowed to remove discrepant data were used. The shape of the travel-time curve and the average station anomalies found are in general agreement with the results of other works (e.g. Herrin et al., 1968) but the scatter of the data is only about half as large (∼0.6 sec), verifying that deep earthquakes are indeed more suitable than shallow ones for travel-time studies. The shape of the travel-time curve is determined within 0.1 to 0.2 sec out to epicentral distances of 70°, but the absolute level of the curve remains uncertain within about ± 1 sec. This “base-line correction” has been estimated from Nevada Test Site explosion data, for which the upper mantle velocity structure near the source is known and could be corrected for. The results of this procedure indicate that the Jeffreys-Bullen (1940) times are about 1.5 sec late but the model of Herrin et al. (1968) has approximately the correct base line of P travel times at least for focal depth of 550 km.
The station anomalies of this study, represented by constants, correlate very well with large-scale tectonic features and also give a better fit to the travel-time data than has been achieved in other studies which included azimuthal station terms.
Systematic regional variations in travel time of 1 sec or more occur beyond 70°, and indicate that significant lateral variation occurs in the lower mantle (depth > 2000 km). A conservative lower bound of about 1 per cent can be placed on the magnitude of these variations in compressional velocity.