Traditional Chinese Medicine (TCM) is the main route of disease control for ancient Chinese. Through thousands of years' development and inheriting, TCM is the most influential traditional medical system which lasts the longest time and used by the largest population. However, the theory of TCM lacks objective and quantitative standards. In this paper, we propose a statistical diagnosis approach to find out the pathogenesises based on the latent semantic analysis of symptoms and the corresponding herbs. We assume that the latent pathogenesis is the inherent connection between symptoms and herbs within a medical case. Previous topic models mostly focus on single content documents, but medical cases have two different contents: symptoms and herbs. We therefore develop a novel muti-content model based on LDA. We used the proposed model to analysis two TCM domains amenorrhea and lung cancer. Experiment results illustrate that the pathogenesises found by our model correspond well with the theory of TCM and it provides a theoretical data-driven approach to establish diagnosis standards.