Tongue diagnosis is a unique method of diagnosis practiced in traditional Chinese medicine and an important tool in diagnosing diseases before determining proper methods of cures and treatment. In the traditional clinical practice, tongue diagnosis depends solely on the personal knowledge and experience of the practitioner, thereby lacking in objectivity and quantification. Consequently, an automatic tongue diagnosis system (ATDS) has been developed to provide practitioners with objective data and assist them in reaching diagnoses. To prove the system¡¦s stability in clinical application, this study employs 20 sets of tongue images, taken 10 minutes apart from 20 patients with possible variations in lighting, and the length, shape and angle of extruding tongue. The features extracted by the ATDS for each set of tongue images are compared to investigate the intra-agreement of ATDS, and two sessions of tongue diagnosis questionnaires to investigate the intra-agreement of doctors with consensus (DC) and doctors without consensus(DWC),and the inter-agreement between ATDS, DC and DWC, and the inter- agreement among DC and DWC. And this study also investigate the improvement of agreement by DWC after a training process.