Weighted K nearest neighbor (WKNN) algorithm provides good result for indoor localization by searching for matching fingerprints in the dataset. However, due to the nature of this method, computation load increases as the size of dataset. In this paper, we propose a hierarchical localization method to reduce the computation load during the online phase. A twolevel structure first localizes the user to a smaller subset and then position estimation is obtained by WKNN with feature extraction. On a public accessed WiFi fingerprinting dataset, the proposed method achieves a 98% reduction without sacrificing localization performance.