Abstract: According to the defect image characteristics on steel strip surface, a defect detection algorithm was proposed on the basis of grayscale projection, where the steel strip image was projected downward and leftward to find out the defects and their positions. The algorithm is easy to implement and the load to be computed is less. The results showed that the average time is about 20 sec to detect a steel strip image by using the proposed algorithm that can eliminate the serious dependence of choosing reference image on the image difference recognition methods and improve further the reliability and stability. The algorithm has succeeded in defect detection on steel strip surface in many experiments.