This paper uses terahertz time-domain spectroscopy (THz-TDS) to analyze the content of seed potato starch in kudzu qualitatively and quantitatively. The spectral data of kudzu powder mixed with seed potato starch was collected, and the qualitative model was established by partial least squares method (PLS) to determine whether the starch powder was mixed with potato powder. The total false positive rate of PLS is 0%, and the model correlation coefficient is 0.925. The results show that PLS can be used to determine whether the starch powder is qualitatively differentiated. PLS and least squares support vector machine (LS-SVM) were used to establish a quantitative prediction model for the content of seed potato starch in kudzu respectively; for PLS prediction model, the determination coefficient is 0.932, and the root mean square error (RMSE) of the predicted set is 2.6%; for LS-SVM prediction model, the determination coefficient is 0.957, and the RMSE of LS-SVM of the predicted set is 1.6%. The results show that the LS-SVM quantitative prediction model is excellent. The research shows that THz-TDS can be used to rapidly and effectively detect the content of seed potato starch in kudzu qualitatively and quantitatively.