An algorithm by combining wavelet decomposition and ranked order mean method (ROM) is provided for the detection and removal of impulse spike noise from measures of track geometry in railway infrastructure. Random events, including anomalous reflection of sunlight, sensors and data transmission errors, always cause spike noise within track measures. Different from the usual random noise, impulsive noise is local noise. So the direct use of the traditional median filtering technology to filter out the spike noise may reduce the useful features and quality of the signal. We develop a new algorithm in which wavelet analysis is applied to decompose the signal into a high frequency signals including spike noise and low frequency signals, and the ROM is proposed to automatically identify the location of the impulsive noise in the high frequency signal and filter it out. Two indicators of ROM have been simplified to one on basis of their inherent relation, and its threshold is determined based on expert knowledge. This method has the advantages of being relatively simple, efficient and robust, and can be achieved online. The method is used for removing the spike noise within analog signal and the measured track alignment irregularity signal, and the results show it is able to not only accurately identify the location of the spike noise and repair the signals, but also retain useful features of large value exceeding and large value alignment within the switches.