Abstract: WiFi fingerprinting localization is currently the most promising method to building large-scale urban localization systems for both indoor and outdoor environments. To reduce the negative effect caused by the fluctuation of the received signal strength (RSS) and improve the positioning accuracy and robustness, an accurate radio fingerprinting localization algorithm based on common beacons is presented in this paper. It formulates the object localization as a Bayesian estimation problem. By employing Gaussian mixture model to accurately represent the complex training fingerprint pattern, and using a Markov-chain state transition model and an adaptive grid collection selection method based on the posterior probability to exploit the object’s past states and environment layout information, the algorithm not only can reduce the size of grid search space, but also can restrict the impossible position jump during the moving process and improves the localization accuracy and robustness. Practical experimental results show that the proposed positioning method can achieve localization error within 3 m with 90 percent probability, which works much better than the single Gaussian localization model.