The paper presents a method which combines the Raman spectrum and the least square support vector machine (LSSVM) based on particle swarm optimization (PSO) to detect the content of three components of edible blend oil rapidly and quantitatively. In this paper, three components of edible oil were investigated. The characteristic peak intensity of Raman spectra was extracted by four pretreatments of the spectra. Then in the training samples, the characteristic peak intensity and the percentage of mixed oil samples were used as the input values and the output values of the regression analysis model. The mathematical models of LSSVM and PSO-LSSVM were established after different pretreatments. The predictive ability of the model was analyzed by the correlation coefficient and mean square error in the test samples. The traditional LSSVM algorithm for nonlinear modeling has many issues, such as its kernel parameter σ and the regularization parameter γ have great influences on the learning model and generalization ability. The fitting precision and generalization ability of the model are dependent on its related parameters, and the time consuming is too long while the optimal step size is too little; however, the global optimal values are hardly to get while the optimization step size is large. Yet, the PSO-LSSVM algorithm has the PSO algorithm advantages of fast convergence and global search capability, which can overcome the problems of time consuming and blindness in LSSVM algorithm. So the kernel parameters σ and γ of LSSVM algorithm are optimized by Global optimization ability and fast convergence characteristics of PSO algorithm. In the quantitative analysis of the three components of edible blend oil, the validation set correlation coefficients of the model for soybean oil, peanut oil and sunflower kernel oil were 0.967 7, 0.997 2 and 0.995 3, respectively; in addition, the mean square errors were 0.054 9, 0.009 2 and 0.047 1, respectively. Compared with the LSSVM algorithm, the prediction accuracy of PSO-LSSVM model is higher and the convergence rate is faster which has been verified by the experiments. Thus, the method can detect the content of the three components of edible oil accurately.