Abstract. In this work, we propose a new scheme for the recognition of document images under a syntactic approach. We present a new method to model the layout of the document using a tree-like representation of the form. The syntactic representation of the documents are used to infer a tree automaton for each one of the classes involved in the task. An error-correcting analysis of tree languages allows us to carry out the classification. The experimentation carried out showed the good behaviour of the approach: error rate of 1.18%.