Ontologies, as the conceptualization of concepts and their interrelations in a specific domain, have proven to be a useful toolin the area of Information Retrieval. However, the relevance of the information they contain depends on their regular updating. Recently, new techniques that automatically build and enrich ontologies have been proposed. A previous work proposed a generic approach that allows any search engine to develop its semantic layer by applying Ontology Learning techniques on Web snippets.This paper applies this idea to PubMed, a well-known medical digital library. The new system (Sem-PubMed) builds automatically new ontology fragments related to the user’s query and then it reformulates queries using the new concepts in order to improve the accuracy of Information Retrieval. The tests of the system have revealed an improvement of the relevance of the search results along with the automatic enrichment of the domain ontology.