Abstract : The human perceptive vision combines several points of view in order to improve the interpretation of a scene. It is modeled by a physiologic component, the perceptive cycle, guided by a psychological aspect, the visual attention. This mechanism is the base of our work on a generic method for document structure recognition. In this context, we propose the formalism of perceptive layer and some multiresolution tools to simulate the perceptive vision and the visual attention. This produces the perceptive method DMOS-P, which is an improvement of the existing DMOS method. Thanks to this method, it becomes possible to easily specify some complex mechanisms of perceptive cooperation, adapted to each kind of document, and that improve the recognition of the structure. We point out a mechanism of prediction/verication, linked to the perceptive vision : at low resolution, hypotheses on the contents are proposed, that are veried at a higher resolution. This mechanism simplies and improves document recognition : for noisy documents, the perceptive vision makes it possible to select only relevant information, whereas for low structured documents, the perceptive vision helps to rebuild the structure. We validated this approach on various kinds of structured documents (incoming mail, archive registers, newspapers. . .), at a large scale (more than 80,000 images) and thanks to an industrial transfer to Evodia company.