Abstract: Using time domain information of the ultrasonic signal can evaluate the quality of spot welding. In the BIW ultrasonic testing process, because of the fuzzy function of the noise signal, the useful information of the welding spot quality is covered up, which leads to the inaccurate result of the spot welding quality evaluation. The ultrasonic signal received by ultrasonic probe can be regarded as the superposition of original ultrasonic signal and noise signal, based on convolution model of the ultrasonic signal and sparse features, combined with minimum entropy deconvolution blind ultrasonic signal convolution, separating overlapping spot welding ultrasonic echo signal and restore the reflection coefficient, to obtain an accurate number of echo. After simulation and experiments, the result proved the effectiveness of the minimum entropy deconvolution on the separation of the overlapped ultrasonic signals.