Several hypotheses have been advanced about the genetic mechanism of back arc basins, but at present none of them is largely accepted. This work aims at recognizing which one of the proposed models may more plausibly be reconciled with the major features of subduction zones and back arc basins in the world and with the results of numerical and analogue modelling of subduction processes. Our analysis points out that the interpretations which explain back arc extension as a side effect of subduction do not provide convincing explanations for some major evidence, as the fact that back arc extension occurs in some subduction zones and not in others, that the same process cessated in zones where subduction has remained active, that the arcs associated with back arc basins are often characterized by a strongly curved shape, that arc-trench-back arc systems do not develop along the entire length of consuming borders and that no significant correlation can be recognized between any parameter of subduction processes and the occurrence of back arc extension. In addition, modelling experiments indicate that the magnitude of the tensional stress induced in the overriding plate by subduction-related forces is significantly lower than the lithospheric strength. These problems are discussed, in particular, for three subduction-related interpretations, the 'slab-pull', the 'corner flow' and the 'sea anchor' models, which seem to be the most quoted in literature. It is then argued that possible solutions of the above problems may be provided by the extrusion model, which postulates that back arc basins are generated by the forced separation of the arc from the overriding plate, along a sector of the consuming border. This separation is generally caused by the oblique indentation of strong and buoyant structures against the accretionary belt. In this view, subduction and back arc extension are not causally linked one to the other, but rather represent simultaneous effects of the lateral migration of the arc. It is pointed out that the conditions required for the occurrence of this kind of mechanism may be recognized in most of the tectonic contexts where back arc basins developed in the wake of arc-trench migrating systems. On the other hand, in the zones where the above boundary conditions are not recognized, as in the South American subduction zones, back arc extension does not occur. It is also suggested that the stop of extension in a number of basins, as the Kurile, Japan, Shikoku, Parece-Vela, Balearic and Pannonian was caused by the interruption of the boundary conditions which determined the deformation of the respective arcs.