The application provides a method and a device for extracting potential area of mechanical equipment fault in an infrared image. The method includes: performing saliency detection on the infrared image to obtain a grayscale image with the background removed; determining a binary threshold of the grayscale image, and according to the binary threshold, converting the grayscale image to a binary image; determining a position of an area whose grayscale value is 0 in the binary image, and extracting an image portion corresponding to the position of the area from the infrared image, as a potentialarea; converting the potential area to a grayscale histogram, extracting feature data of the grayscale histogram, and according to a feature data training set, training a preset classifier to obtaina fault classifier; testing whether prediction accuracy of the fault classifier reaches a preset accuracy rate; If the preset accuracy rate is not reached, obtaining a new binary threshold and performing an iterative loop calculation until a preset iteration termination condition is met, and then re-determining a potential area according to the corresponding binary threshold. According to the application, a more accurate potential area of fault can be extracted.