Abstract: Hardware Trojan, a malicious circuit inserted into the golden circuit, causes a serious risk to the security of Integrated Circuits and the trust of critical system. Thus, hardware Trojan detection is of great significance. A novel hardware Trojan detection method based on clustering analysis algorithm is presented in this work, which employs the global optimized ability of genetic algorithm and the local searching capability of K-means algorithm to automatically fulfill the detection online. Side-channel signals, such as global clock signal, reset signal and a ring oscillator signal of a chip, are applied as the inputs of the clustering algorithm, since they may be altered by the Trojan circuit. Experimental results implemented with FPGAs demonstrate that the tested chips can be clustered into two categories, and the Trojan chips can be distinguished from the Golden chips accurately. The detection resolution achieves about 10^-4, which indicates the proposed novel detection method is feasible and effective.