Tongue diagnosis with features like tongue coating, petechia, color, size and so on is of great effectiveness and convenience in traditional Chinese medicine. With the development of image processing techniques, automatic image processing can reduce hospital inspection for patients. However, there are ubiquitous problems of inadequate accuracy in petechia dots detection with previous methods. In this paper, we propose a method of petechia dots detection on tongue based on SimpleBlobDetector function in OpenCV library and support vector machines model, which improves the detective accuracy. We test 128 clinic tongue images and select 9 of the images with plentiful petechia dots for further experiments. Our method achieves mean value of false alarm rate 4.6% and missing alarm rate 11.8%, which have 19.4% and 8.2% reduction respectively compared to previous work.Clinical Relevance-The method can provide detailed information of tongue, which assists doctors to investigate curative effect.