Knowing the location of a train is necessary to develop a useful service for train passengers. However, popular localization methods such as GPS and Wi-Fi are not accurate especially on a subway. As an alternative method, estimation of motion state and stop station by using sensors on a smartphone is being studied for subway passengers. This paper proposes a localization method that uses only a barometer on a smartphone. We estimate motion state from the change of elevation, and also estimate latest stop station by the similarity of a series of elevations recorded when the train stopped and actual elevations of stations. By estimation of the motion state and the latest stop station, development of various context-aware services for subway passengers becomes possible. Through experiments in four lines of subway in Tokyo, we demonstrated that the accuracy of estimation of the motion state is 86%, and estimation of the stop station is 58%.