Abstract: The commonly used scintillator detectors are sensitive to neutron and γ ray, and it is necessary to eliminate the influence of γ ray on neutron measurement. The support vector machine (SVM) can realize the function of two classifiers. In this paper, the SVM combining principal component analysis (PCA) and genetic algorithm (GA) was applied to particle discrimination in n/γ mixed field. PCA was used to reduce the dimensionality of eigenvalue to avoid SVM over-fitting. At the same time, GA found the optimal value of SVM key parameter penalty factor C and kernel function parameter g through iteration. The accuracy of PCA-GA-SVM in the discrimination of n/γ pulse waveform was verified and compared with the discrimination result of charge comparison method and frequency gradient analysis method. The results show that the discrimination accuracy of the SVM network optimized by PCA and GA is improved significantly. This method can provide effective discrimination for n/γ mixed field.