Abstract: Aiming at the insufficient accuracy of the template matching methods and the defect detection methods used in PCB defect detection system, a multi-level detection method for PCB board defect based on improved template matching and image difference was proposed. Firstly, an improved template matching method based on multi-direction Sobel operator was proposed to realize the matching between the tested image and the template image. On this basis, a multi-level binaryzation detection method based on image difference was proposed to realize the defect detection of PCB image. Combined with the results on the experimental platform, the proposed method can extract the defect area accurately. The false detection rate of this method is only 5%, which has better stability and higher accuracy.