Under the background that conveyor belts are prone to rips in operation, According to the characteristics of human visual system and information theory, an algorithm for detecting rips in conveyor belts based on visual saliency and maximal entropy is proposed. Regions of interest are firstly extracted in frequency domain, and then a threshold is obtained through calculating the maximal entropy of saliency map to segment the grayscale image. Experiments show that the algorithm is stable and adaptable.