Based on complexion-viscera diagram in TCM (Traditional Chinese Medicine), a method for complexion recognition was proposed for the first time. Firstly, the sampling environment was standardized, and 68 common facial feature points were localized by AdaBoost and ASM algorithm. Second, the complexion features in LAB color space were extracted from 15 diagnostic feature points on complexion-viscera diagram by FCM clustering which separates the complexion feature from skin color. Finally, the complexion features were classified by SVM algorithm. This method can be applied to the automatic logic inference of face diagnosis in TCM. Experiments on a test set of 400 face images result an accuracy of 84.6%. Results prove the effectiveness and efficiency of this method. © 2010 IEEE.