Reliable depth imaging of geological structures has always been a crucial point for the oil industry. For complex structures, the problem is far from solved, the main difficulty being the determination of the velocity model. We have adopted a kinematic approach for this determination, i.e. an approach which explicitly uses and relies on travel-time information. Such information is provided by an interpretation of prestack seismic data. First of all, the data are interpreted in order to obtain information on the structural depth model and on the velocity model of the geological structure under consideration and to acquire kinematic information (i.e. traveltimes) about the subsurface. However, this last information is likely to be rather sparse. Reflections from deep reflectors often cannot be identified because of wave propagation effects through the complex overburden. We therefore have to increase our information. We propose to do this by use of a recursive scheme. The available information is processed by tomographic inversion in order to obtain initial estimates of both a structural and a velocity model of the subsurface that match the kinematic information in hand. Moreover, the solution is constrained to match such a priori geological information as may be available. To extract more information from the data, we now make use of prestack depth migration. Although the velocity model provided by tomographic inversion may be far from perfect, the migration results obtained will make some events appear that were barely visible on the original data. In addition, these events can be followed from one migrated common-shot (or common-offset) gather to another. Migration coherency panels make it possible to interpret additional reflector elements. Kinematic modelling of the associated refletions yields new traveltime information which is then used in the next iteration of the tomographic inversion. This closes the recursion loop. A significant point is that new, correct kinematic information can be obtained even with an erroneous velocity model. Migration and modelling are, in fact, mutually inverse processes kinematically speaking. Modelling undoes what migration did. Another feature of importance in the migration-aided tomography method is that the insight gained from interpreting partial results can be fed into the process.