Abstract: The automated essay scoring for the teaching of college English writing requires that the scoring method should have the feature of generality，namely，without pertinency of specific subjects.In the aspect of content evaluation，document clustering can put eassys together according to the similarity of their contents to form a clustering tree which has a higher similarity in the core than in the peripheral area of the tree.A few essays that locate in the peripheral area are quite different from most others in content.These essays are possibly off the topic and will be submitted to teachers for further examination. By this way，eassy contents can be evaluated accurately with only minor labor expense.Experiment shows that this method can identify essays off the topic effectively with a reasonable threshold value of content similarity.