The production of abstracts of documents is an interesting process: a source text is read and understood to recognize its content which is then compiled in a concise text. The idea of producing abstracts automatically is not new but the results obtained until now do not meet the demands of the intended audience. We are interested in the production of abstracts using natural language generation theories and techniques which have been, until now, mostly ignored in automatic abstracting. In this talk we will present several issues involved in the process of abstracting.