Although social tagging systems are becoming increasingly popular, tagging is still usually a manual process. When publishing on a social tagging system, the user is asked for the tags he wishes to assign to the resource being made available. In this paper, we present an automatic tag suggester, Tess. Our system makes recommendations based only on the textual contents of the resource and is independent of existing tags, thus allowing the emergence of novel tags. The system was evaluated by a group of users and statistical measures were applied to infer its performance. Results show that the system is not only able to suggest many useful tags, but also to discover new and relevant tags, not suggested by any of the human users.