To construct a Traditional Chinese Medicine(TCM) differentiation model by taking qi deficiency syndrome differentiation model as an example. Methods: To input the 684 cases and medication records of TCM distinguished veteran doctors into the electronic medical records management system and build a table of symptoms using its statistical function. Then using BP neutral network to input the symptoms table into Matlab software and take 669 medical records as training data randomly. The last 15 medical records are taken as test data. Results: The absolute error in test data and model data shows that there are 3 cases that are greater than 0.6, and the other 12 cases were less than 0.3. The accuracy is 83.3％, specificity is 77.8%, and predication consistency is 80%. Conclusion: This research has developed a TCM syndrome differentiation model with high accuracy based on BP neural network, and explored a new way to summarize experience of TCM distinguished veteran doctors. Therefore, it is worthy of popularizing.