A damage detection method for transmission tower is presented based on integrating the cross correlation function amplitude and the support vector machine (SVM). The proposed method consists of two stages. Firstly, the data features, which are defined as the difference of the cross correlative function amplitude between initial and current states, is calculated from the approximate signal of single mode responses. The data features are then used to train the SVM classifier to turn the damage identification problem into classification problem. The feasibility of the method is verified by the vibration test of the transmission tower model. The results show that the proposed method only uses the dynamic responses from small number of sensors, and can identify the damage status with strong noise robustness during arbitrary excitations.