In this paper. we propose a new method for automatic classification of documents using the degree of similarity between words. First, we seek relevance terms using automatic indexing. Second, we found frequency in use words in documents and the degree of relevance between the words using probability model. Continuously, we extracted the set of words which is connected the relevance closely and created the profiles characterizing each classification And, with the profile we finally classified them. We experimented on classifying two groups of documents. Some documents were about Genetic Algorithm. The others were about Neural Network. The results of the experiments indicated that automatic classification with word accordance of degree enable us to manage the retrieved documents structurally.