Abstract: Based on the finite element simulation models of spot welding ultrasonic detection for low carbon steel and Fast Fourier Transform, the frequency spectrums of four kinds of simulated signals, such as pore, deep indentation, small nugget, loose weld, were obtained. And the time domain and frequency domain features of signals were extracted respectively by statistical methods. By analyzing the propagation rules of ultrasonic in different defect spot welds and the change rules of features, the recognition methods of different defects types of spot welding were summarized. The method was used to identify the defects of a large number of spot welding samples by ultrasonic testing signals, and the results were compared with the metallographic test results. The results showed that the time domain and frequency domain features of ultrasonic detection signals were effectively recognized the types of spot welding defects.