Abstract: Aiming at the problem of parameter selection and time overhead for BP neural network and SVM algorithms, this paper proposes a Hardware Trojan classification method based on Random Forests. Firstly, the Hardware Trojan detection problem is modeled as a binary classification problem and the power consumption of the chip is sampled several times. Then the characteristics of the power consumption curve are extracted by the PCA (principal component analysis). Finally, RF (Random Forests) classification model is used to classify the feature vectors in purpose of identifying Hardware Trojan chips. The experimental results show that, considering the same Hardware Trojan horse data processed by PCA, the discrimination accuracy of RF is improved by 9.13% compared with the BP neural network. Compared with the SVM(support vector machine) method, the discrimination accuracy is increased by 15.96%. Compared to the other two methods, the time cost of RF is reduced by about 8 times.