New empirical traveltime curves for the major seismic phases have been derived from the catalogues of the International Seismological Centre by relocating events by using P readings, depth phases and the iasp91 traveltimes, and then re-associating phase picks. A smoothed set of traveltime tables is extracted by a robust procedure which gives estimates of the variance of the traveltimes for each phase branch. This set of smoothed empirical times is then used to construct a range of radial velocity profiles, which are assessed against a number of different measures of the level of fit between the empirical times and the predictions of the models. These measures are constructed from weighted sums of L2 misfits for individual phases. The weights are chosen to provide a measure of the probable reliability of the picks for the different phases.
A preferred model, ak135, is proposed which gives a significantly better fit to a broad range of phases than is provided by the iasp91 and sp6 models. The differences in velocity between ak135 and these models are generally quite small except at the boundary of the inner core, where reduced velocity gradients are needed to achieve satisfactory performance for PKP differential time data.
The potential resolution of velocity structure has been assessed with the aid of a non-linear search procedure in which 5000 models have been generated in bounds about ak135. Misfit calculations are performed for each of the phases in the empirical traveltime sets, and the models are then sorted using different overall measures of misfit. The best 100 models for each criterion are displayed in a model density plot which indicates the consistency of the different models. The interaction of information from different phases can be analysed by comparing the different misfit measures. Structure in the mantle is well resolved except at the base, and ak135 provides a good representation of core velocities.