Abstract: The main challenges of indoor localization come from multi-path propagation and non-stationary channel environment. Some classical localization approaches based on single received signal strength (RSS) fingerprint show low accuracy and bad robustness due to some environment changes. In this paper, we propose an accurate indoor localization algorithm by fusing group of fingerprints via Dampster-Shafer (D-S) evidence theory. The main idea can be summarized as follows:in off-line phase, first, based on the received data from a receiving array deployed in indoor environment, we calculate four fingerprints, namely, RSS, covariance matrix, signal subspace, and fourth-order cumulant. Secondly, these fingerprints are input to train four different classifiers by using back-propagation (BP) neural networks. In on-line phase, by calculating the corresponding transformations of the received signals of the array, we can obtain the predictions of these classifiers; then, we use D-S evidence theory to fuse the final localization results. The proposed algorithm can deal with different environment noise adaptively and show higher accuracy compared with some existing fingerprint-based algorithms. The performance of our proposed algorithm is verified by simulation results.