To establish a Bayesian diagnosis model of TCM symptoms by using the hypertension epidemiological Syndrome databases. The Clementine 12.0 software is used to build the Tree Augmented Naive Bayes models, calculate the Bayesian conditional probability, and compare the forecast accuracy of the syndrome diagnosis model. The training sample had 384 cases and calculated 69 symptoms and signs, without prior knowledge, the prediction accuracy rate of the training model are 72.11%, and with the prior knowledge, the testing sample had 384 cases, the prediction accuracy rate of testing model is up to 78.55%. Through the sample study, Bayesian networks can improve the prediction accuracy; we can build a more accurate hypertension diagnosis model through the current work.