The current corpora of question classification are relatively small and difficult to meet the practical needs of Question Answering system, so that we use active learning methods to construct a Chinese question classification dataset and for question labeling. In addition, we improve the performance of labeling with fea- ture selection. Experimental results show that by using active learning we can quickly converge at the best ac- curacy (85% ) and by using manual tagging we can have small feature set size. The active learning-based la- beling method achieved very good classification performance with less manual annotation tagging, which can significantly improve the accuracy of classification to some degree