Wifi fingerprinting using received signal strength has been widely studied for indoor localization. Classic similaritybased methods like weighted K-nearest neighbor (WKNN) localize targets by searching the best matching fingerprint in the dataset. Performance of these methods suffers from RSS variance and they are slow under a large size of fingerprint dataset. In this paper, we propose a WKNN localization strategy using k-means clustering radio mapping to improve localization precision while mitigating computational complexity.