Indoor localization is becoming increasingly important for mobileapplications. WLAN fingerprinting is a compelling technique becauseit builds upon existing infrastructure and client hardware availablein off-the-shelf mobile devices. We evaluate different methods forWLAN fingerprint classification with a focus on on-device localization.The main scientific contribution of this approach is that any Androidbased device can localize itself (without any server being able todetermine the current location) using existing WLAN infrastructure(no additional access points have to be installed, the firmware ofexisting access points doesn’t have to be changed). This approachwas chosen to make indoor localization feasible in non-academic usecases.With a functional implementation and a simple procedure for collectingWLAN fingerprints, we currently achieve an accuracy of 4,m in 90%of all cases with a mean error of only 2.2,m when the same deviceis used for training and testing. Next steps are calibration betweendifferent mobile devices, post-processing in terms of movement, andautomatic downloading of the required WLAN fingerprint databaseson a global scale.