This article presents a system to extract Knowledge from webpages by producing semantic annotations. taking into account semantic information from the domain to annotate an element in a webpage implies solving two problems : (1) identifying the syntactic structure of this element in the webpage and (2) identifying the most specific concept (in terms of subsumption) of the ontology that will be used to annotate this element. Our approach relies on a wrapper-based machine learning algorithm combined with reasoning making use of the formal structure of the ontology.