Abstract: This study presents a method on automatic question classification through attribute and value reduction based on rough set theory. The core of the method is adopting statistical machine learning, with the assistance of a fair number of training corpus, attempts to automatically obtain useful and concise classification rules. Attributes reduction algorithm can omit the attributes which are unnecessary to decision classification in the decision table so as to simplify the decision table and increase the adaptability of decision process. The value reduction algorithm based on attributes significance can further eliminate the unnecessary information in the decision table. Comparing with the alternative means under the same data set and classification architecture, the experiment result is that the accuracy of the rough classification in this study is up to 86.20%, fine classification reaches 78.8%. It means that the method of this study is efficient.