Spectral analysis is an important means of studying the interaction between THz radiation and matter. The transmittance spectra of samples with different levels of saccharin are tested using an all-fiber THz-TDS system, and it is found that the characteristic absorption peaks of saccharin are located around 1.4 THz and 1.7 THz. PCA-SVM method is used to establish the regression model of saccharin content, and the prediction results are analyzed and compared with the GA-PLS model. Correlation coefficients and RMSE are introduced to evaluate the modeling effect, and the sample set made with a 10% content gradient is tested. The research results show that the RMSE of the prediction models established using PCA-SVM, SVM and GA-PLS methods are 1.885%, 1.926% and 2.432%, respectively. Therefore, the PCA-SVM method has the best prediction performance, and the predicted data show a good correlation with the actual data. A content regression prediction model with good performance is obtained, which provides an effective means for the detection and analysis of saccharin content.