Abstract : This paper deals with thecategorisation of textual cues in scientific abstracts with the aim to highlight the information contained while for exploring huge volumes of texts. Typically, one context of application is the rapid identification by an expert of strategic information for science and technology watch. From a study of a sample of abstracts in english, novelty, objective, result and conclusion cues are formalized as finite state automata and projected on a test corpus. Resultats show that using these cues is relevant. Using the type of cues identified and the supposed information announced, an XML markup of scientific abstracts is proposed. The final aim is to guide the reader towards information categories classified as such which can assist science and technology watch process.