In inductive learning, to build decision trees is often arduous when there exists more than two classes to learn. In this paper, a method of decomposition of problems with more than two classes into several problems with only two classes is proposed. This decomposition enables the construction of a forest of fuzzy decision trees where each fuzzy decision tree is dedicated to the recognition of a single class against a combination of all the other classes. The construction of fuzzy decision trees is based on an extension of the ID3 algorithm which handles imprecision in data. A method to use such a forest of fuzzy decision trees to classify new cases is also proposed.