As China's railway transport system is getting busier and the trains are continually speeding up, the rails have to bear more and more heavy cyclic loads generated by the repeat force between the wheels and rails, and the internal damages are easier to be produced, which directly endanger the running safety. Therefore, it is badly in need of rail flaw detection equipment with high efficiency, more intelligence and low cost. A new rail damage detection car based on ultrasonic testing technology is designed and a series of digital signal processing algorithms are applied to estimate the position of defect in this paper. The position of defect is estimated by ultrasonic time-of-flight (TOF). Firstly, over-complete dictionary of atoms is founded by simulated model for damage echo signal. Then the ultrasonic echo wave is decomposed by matching pursuit technology with over-complete dictionary of atoms after noise reduction by the wavelet transform. Finally, the TOF is obtained by cross-correlation algorithm. The result of experiment shows that the rail detection car and the method to estimate the TOF are experimentally feasible and they can meet requirements of rail flaw inspection.