Mining frequent subtrees from tree databases is currently a very active research topic. This research has many interests, including for instance mediator schema building from XML schemas. In this framework, many works have been proposed. However, the way the data are represented is often very memory-consuming. In this paper, we propose a new method to represent such data. We show that this representation has many properties , which can be used in order to enhance sub-tree mining algorithms. Experiments are led to assess our proposition (data representation and its associated algorithms).