The reconstruction of the western Mediterranean since the Oligocene emphasises the role of subduction rollback in convergent plate margins. It shows that a widespread extension took place in the convergent interface separating Africa and Europe, forming marine back-arc basins and new spreading centres. Subduction rollback was probably controlled by the gravitational instability produced during subduction of cold and dense oceanic lithosphere. Back arc extension was likely to occur when the velocity of the slab retreat overcame the absolute motion of the overriding plate (Molnar & Atwater 1978, Dewey 1980, Royden 1993b, Lonergan & White 1997). In the evolution of the western Mediterranean these processes have played a fundamental role since the Oligocene. Subduction rollback was made possible by the consumption of Mesozoic oceanic lithosphere at the subduction zone. This oceanic lithosphere was probably old and cold enough to be gravitationally unstable relative to the surrounding asthenosphere. The subduction zone has therefore progressively retreated from its Oligocene position near the southern margin of Europe to its final configuration in the Calabrian arc, North Africa and the Alboran arc. In the western Mediterranean, subduction rollback occurred during a period of relatively slow convergence between Africa and Europe (Jolivet & Faccenna 2000, Rosenbaum et al. in press). As convergence alone could not compensate the vacant area formed by subduction retreat, back-arc extension occurred in the overriding plate. Thus, a period of relatively slow convergence was actually characterised by large-scale horizontal motions of smaller microplates and allochthonous terranes.

The dispersion of continental terranes, which drifted and rotated during subduction rollback, is clearly seen in our reconstruction. In the Alpine orogen, this process led to the fragmentation of a continuous belt into continental terranes, which in turn, collided with the passive margins of the surrounding continents. We stress that this mechanism may have profound tectonic implications on the way orogens work. Orogens may be subjected to switches from crustal shortening and extension, controlled by the processes of subduction rollback, rifting in the back-arc region and the subsequent accretion of allochthonous terranes onto adjacent passive margins. Thus orogenesis cannot be oversimplified to subduction followed by collision of two continental plates, but includes accretion of numerous continental terranes. Following collisional events, reorganisation of the plate boundaries occurs, associated with termination, jumping or segmentation of subduction zones. A Similar style of orogenesis has been proposed by Nur and Ben Avraham (1982) based on numerous examples of allochthonous terranes throughout the Circum-Pacific (excluding the Andes) and the Alpine Himalayan belts. These authors have suggested that continental slivers and microcontinents could actually migrate great distances before colliding with the continents.

In summary, the style of tectonism suggests that the role of horizontal motions during orogenesis cannot be disparaged. Fragments of continental crust were possibly subjected to large amount of horizontal transportation, block rotations on vertical axes, and episodic alterations from crustal shortening to extension.