The classification of concept-drifting data streams with complex category structures has recently becomes one of the most popular topics in data mining. This paper proposes a novel subspace classification method, and uses it to form an ensemble classifier in a hierarchical structure for concept-drifting data streams classification. After dividing a given data stream into several data blocks, it uses the subspace classification method to train some bottom classifiers on each data block, and then uses these bottom classifiers to form a base classifier. The base classifers are used to build the ensemble classifier. Meanwhile, it introduces the parameter estimation method to detect concept drift. Experimental results show that the proposed method does not only significantly improve the classification performance on datasets with complex category structures, but also quickly adapts to the situation of concept drift.