A damage detection method based on wavelet packet transform and replicator neural networks is proposed to carry out real-time damage detection of structures. Firstly,wavelet packet transform is used to decompose original vibration response signals to calculate the relative band energy of each band. The distribution of relative band energy reflects the characteristics of the structure to be detected. Secondly,the relative energy from healthy structure is taken as input to training replicator neural networks. Finally,the trained networks are used for damage detection in real-time. Experiments showed that the proposed method can detect single-position damage even under noise interference.