In this paper, a novel and simple method is employed to automatically construct domain knowledge base for query classification from large-scale Web pages. Besides, using context as the feature of words, the resource of relevant words is built automatically in order to extend the user's query. On the basis of domain knowledge base and extension of the query using relevant words, satisfactory performance in query classification is achieved. Experimental results demonstrate that our method achieves precision of 77.68% and recall of 75.34% in Chinese query classification. In English experiments, in spite of the scarcity of English Web pages and absence of stemming, precision achieves 58.83% and recall achieves 54.13%, which is a great improvement compared to state-of-the-art query classification algorithms.