The thermal structure and evolution of the oceanic lithosphere is revisited with the help of a global shear velocity model of the upper mantle. Seismic velocities of the lithospheric mantle are converted to temperatures, with a particular care dedicated to the estimation of final uncertainties. These are evaluated from a Monte Carlo error propagation, taking into account the uncertainties on velocities and those on the parameters related to the velocity–temperature relationship (mantle composition, thermoelastic properties and attenuation factor). The seismically derived temperature, averaged by age interval, serves to constrain the thermal structure of the lithosphere, together with surface heat flow and ocean depth. Using an experimentally determined thermal expansivity α, the Chablis model, which prescribes a constant heat flow at some isotherms, provides a much better fit to all data than the plate model, which imposes a constant basal temperature. Only a strongly reduced α (30% reduction) allows the latter model to achieve a joint fitting comparable to the Chablis model, and then with a fit to seismically derived temperature that remains inferior the latter model. The good fit of the plate model thus depends on a reduction of α down to the lower possible limit and relies mostly on ocean depth, whose behavior at old ages is considerably obscured by anomalous crust. The Chablis model therefore appears favored by this study, which should give new perspectives on various processes related to mantle convection and to the dynamics of the lithosphere.