Thyroid nodule is a common clinical condition. Ultrasound is usually used to make a preliminary diagnosis, because it is convenient and cheap. Therefore, the study of thyroid ultrasound images of thyroid nodules has it’s significance and value. This paper investigates the problem of locating thyroid nodules in ultrasound images by manual signs. The solution to this problem is divided into three parts: first, image processing processes the image preliminary and find the approximate location of the signs; then, sign recognition recognize the signs accurately using CNN models; finally, boundary adjustment is used for the final adjustment of the border. Experimental results show that the algorithm proposed in this paper can accurately locate the nodules in thyroid ultrasound images.