When performing document search, scientists have specific goals in mind. We conducted interviews with scientists to understand exactly how they were looking for information and working with documents. We found that scientists are generally searching specific discourse elements, not the en- tire document. Therefore, we created an annotation model that can represent the different types of discourse elements contained in documents. We have implemented this model in the form of an OWL ontology and a semantic indexing and retrieval tool. The experiments we have conducted (in the gender studies field) show that the model is sufficient to represent a large part of the document contents and that it is possible to automatically annote documents according to this model. We also showed that this model can be used to answer specific and complex queries on a corpus of scientific documents.