Geological problems raised by the interpretation of deep seismic reflections in collision belts result from the structural complexity of the crust and the lack of surface control and well-documented references. Some of these problems are illustrated by examples coming from various tectonic settings in France and adjacent areas, and tentative interpretations are proposed. From field, conventional seismic and borehole data, interpretation of shallow structures such as thin-skinned detachments is fairly well controlled down to 5 s TWT, but deeper reflections are more difficult to interpret. The shallow detachments are linked to mid-crustal dipping reflections that are regarded as ramps, which may root into deeper intracrustal detachments. Dipping reflections may sometimes develop in the whole crust and produce puzzling criss-crossing patterns. The Wind River thrust provides a well-documented analogue' of such features, although it is located in the foreland of the belt. It suggests that this kind of feature was formed in a late stage of the tectonic development. The lack of any seismic image of the external thin-skinned detachment in large parts of the Palaeozoic belts of Europe can be explained by such a late deformation which affected the belts up to the suture areas. Interwedging patterns ('crocodile') are commonly associated with shallow ramp anticlines. Though hard to establish in the deep crust, interwedging may be present in collision belts as the result of lithosphere punching (e.g. Pyrenees) and/or reactivation of former antithetic shear zones. Steep faults, unambiguously cutting across the whole crust, are scarcely observed and seem to be restricted to areas where major transcurrent displacements are in progress. Dipping reflections across the deep crust and the upper mantle may be interpreted as the ghosts of former crustal slabs now largely metamorphosed in eclogites.