The objective of the thesis is to propose a data-processing model to represent, build and exploit textualstructures. The suggested model relies on a «type/token» form of text representation extended bysystems of lexical and contextual annotations. This model's establishment was carried out in the SATOsoftware -- of which the functionalities and the internal organization are presented. Reference to anumber of works give an account of the development and use of the software in various contexts.The formal assumption of the textual and discursive structures find an ally in the beaconing XMLlanguage and the proposals of the Text Encoding Initiative (TEI). Formally, the structures built on thetextual segments correspond to graphs. In a development driven textual analysis context, these graphsare multiple and partially deployed. Their resolution, within the fastening of the nodes to textualsegments or that of other graphs, is a dynamic process which can be sustained by various dataprocessingmechanisms. Examples drawn from textual linguistics are used to illustrate the principles ofstructural annotation. Prospective considerations for the data-processing establishment of amanagement system of the structural annotation are also exposed.