A Computer-assisted Non-invasive Diabetes Mellitus Detection System through facial key block analysis is designed and developed in this paper. There are four main steps in our system: facial image capture through a non-invasive device, automatic location of the key blocks based on the positions of the two pupils, key block texture feature extraction using Local Binary Pattern with cell-size 21, and classification with Support Vector Machines. In the first step of this system, a specially designed facial image capture device has been developed to capture the facial image of each patient in a standard designed environment. According to Traditional Chinese Medicine theory, various facial regions can reflect the health status of different inner organs. Based on this, four key blocks are located automatically using the positions of the two pupils and used in Diabetes Mellitus detection instead of employing the whole facial image. For the last two steps, an experiment which selects the best value of Local Binary Pattern cell-size and the better classifier of two traditional classifiers (k-Nearest Neighbors and Support Vector Machines) is implemented and its results are applied in this system. In order to test the system performance, the facial images of 200 volunteers consisting of 100 Diabetes Mellitus patients and 100 healthy persons are captured and analyzed through this system. Based on the test result, the Computer-assisted Non-invasive Diabetes Mellitus Detection System through facial key block analysis is proven to be effective and efficient at distinguishing Diabetes Mellitus from Healthy patients in real time.