Recently, the discovery of deep Web data source and domain-relevant issue attract more and more attentions. This paper proposed a method using multi-classifier to discover and classify the data source of deep Web. Firstly, It used naive Bayes classifier to class the page into domain relevance or not. Secondly, improved C4.5 decision tree algorithm was used to identify the query interface. The result of the experiment competed with single decision tree classifier proved this method is effective.