Shear wave travel time residuals have been measured for 25 oceanic earthquakes, including normal faulting events on crests and intraplate events. When the residuals are plotted against the age of the lithosphere in which the event occurred, they show a decrease of 6¿1 s from ridge crest to oceanic lithosphere 100 m.y. old. This decrease is caused principally by an increase in shear velocity within the upper mantle as a function of lithospheric age. Study of depth phases from oceanic earthquakes indicates that all such events are shallower than 10 km below the sea floor, so focal depth variations are not an important contributor to the travel time residual versus age relationship. Modeling of seismic velocities in the upper mantle, using a standard thermal model for the spreading lithosphere and published phase diagrams for dry and wet gabbro-eclogite and peridotite, was conducted in an attempt to produce a residual curve similar to the observed. Comparison of the predicted and observed residuals indicates that the mantle cannot be dry but is unable to distinguish between the wet systems, principally because of the large number of assumption involved in modeling velocity changes across phase boundaries. However, the wet gabbro-eclogite models is unable to match local shear velocities in the oceanic lithosphere obtained from Sn wave velocities and inversion of surface wave phase velocity, and a peridotitic upper mantle is favored. A strong azimuthal variation of residual exists for an event along the Gibbs fracture zone in approximate agreement with the velocity versus age models. Shear wave residuals for events on the Indian plate are systematically less by 2 s than residuals for earthquakes in lithosphere of comparable age in the Atlantic or Pacific; this offset may result from the rapid northward motion of the Carlsberg and southeast Indian ridges with respect to the mesosphere.