We use International Seismological Centre data between the years 1964 and 1994 to derive a data set of P and S station corrections. By restricting the raw P and S data to those which share the same source to receiver paths, we minimize any bias to the station corrections from differential P and S data distributions. We find that both static and azimuth dependent terms are highly correlated. In addition, we perform synthetic tests to determine the sensitivity of our data to mantle structure and show that static terms are sensitive to structure, at least throughout the upper mantle. In determining the regression slope of the static terms, we allow for regional offsets and find a best-fit global slope of S to P correction a=2.85¿0.19. This is lower than that obtained in previous station correction studies but is in agreement with results in the upper mantle using differential travel times and with values derived for the upper parts of the lower mantle. Regionalizing the data by tectonics does not give statistically distinct ratios of S to P station correction. The result for shield areas, however, is significantly smaller than previously believed, probably because of our ability to derive station corrections from similar data sets. Variations in the second azimuthal terms are similar in both fast direction and relative magnitude. In another synthetic experiment we find that their general trends can be predicted well by isotropic models of heterogeneous velocity structure. From this result and also because they do not correlate with SKS splitting data, we conclude that station correction second azimuthal terms should not be interpreted in terms of anisotropy. ¿ 1997 American Geophysical Union