The present article is dedicated to railway track diagnostics using inertial modules based on micromechanical sensors, which are mounted directly on the axle boxes (end bells) of the wheel pairs. A prominent problem in this field is the fact that tread contact surface defects are present not only on the rails being diagnosed, but also on the wheels of the flaw detector car, the axle boxes of which the sensors are installed on. Before the present work, attempts to properly separate the signals caused by these two sources were unsuccessful. The current article describes the authors' algorithm which allows to separate these signals. The solution is based on coprocessing of data acquired from two inertial modules installed on axle boxes of successively travelling wheels. This allowed to increase the reliability of detecting surface flaws by about 10%, and also created prerequisites for improving the accuracy of the travelled distance determined by the odometer.