Abstract: In this paper, an acoustic acquisition system has been developed based on LabVIEW to sample process signals from the railway flash welder. The obtained data was then contrastively analyzed by four kinds of time-frequency analysis method after being denoised. Thus, the STFT time-frequency analysis method was found out to be the best analysis method for acoustic signals during flash welding. On the basis of this research, the weighted average frequency and the average power spectrum values of flash acoustic signals were regarded as the weld quality characteristic values and then an algorithm was developed to explore the relationship between these two parameters, with the help of large data analysis of 26 joints. After contrastively analysis of the characteristic values and the grey-spot area of joints, the relationship between the weighted average frequency, the average power spectrum values of the flash acoustic signal and the Grey-spot area was calculated. This research shows that the low grey-spot area joints have two characteristics in stages of low voltage flashing and the accelerated flashing: the average power spectrum value experiences a steady increase and has no abrupt change up to 200%, and the weighted average frequency experiences a smooth rise up to 20%, especially in the accelerated flashing stage.