In the framework of the Semantic Web, content-based processing of data is considered as an essential component for application semantic interoperability. In order to enable such a processing, many W3C standardisation proposals (rdf, rdfschema, owl, etc.) were done. In this paper, we present an approach for semantic similarity. It is based on content-based annotation of textual documents. The contents of documents are represented by labelled trees where nodes are represented by concepts in an ontology which represents the knowledge of the domain of data. Then, a reasoning process is carried out for comparing the labelled trees representing documents and thus comparing the documents. This comparison process, once completed, allows to calculate a semantic similarity measure between documents. Finally, we show how this semantic measure can be used to classify documents according to their content.