To the shortage of the traditional analysis methods about train impact, this paper put forward a new method using autocorrelation theory and virtual instrument technology to analyze train impulse. Using a double-MCU system, the acceleration signals were acquired at different speed by train, and were transmitted into PC through USB interface. Besides the impulse signals, the acquisition data included other useless signals. The autocorrelation function was small when trains run steadily, but was greater during train impact happened. So the autocorrelation function was adopted to distill the valid impulse data. After frequency domain analyzed and autocorrelation analyzed on the Virtual Instrument flat, a new train impulse grade assessment criterion was built, based on the correlation peak and the width of the peak. In experiment, the impulse signal was separated from noise signal well and truly, and the quantitative model of evaluating train impulse was believable. This system possessed a certain extent theory value and application value.