L-Phenylalanine and L-Tyrosineplay essential roles in synthesizing neurotransmitters and hormones. The two amino acids have similar structures which lead to an obviously functional distinction between the two amino acids. Previous studies have shown that there are remarkable differences between the two amino acids on low-frequency vibrations. Recently, terahertz (THz) spectroscopy has been proven to be a useful technique on studying low-frequency dynamic of biologic molecules. Many multivariate calibration methods have been successfully applied to quantitative analysis multi-components spectra data due to the linear behaviors revealed by terahertz absorption spectra. However, the predictive performances of traditional calibration techniques are sometimes unsatisfied as only a single model is built between spectra and targets to predict the unknown samples. Thus, the ensemble modeling method with better accuracy came into being. The empirical mode decomposition (EMD) method, firstly proposed by Dr. Huang in 1998, is used to decompose the signal into a set of intrinsic mode functions (IMF) self-adaptively, which is widely applied in signal and spectra processing. We proposed an empirical mode decomposition (EMD) based partial least squares (PLS) method for terahertz spectra quantitative analysis on amino acids mixture with various concentrations. The terahertz time signals were decomposed into a series of intrinsic mode functions (IMF) with different frequencies by the EMD method. The several top IMFs (from 2 to 5) based absorption spectra were obtained for quantitative analysis by employing PLS. The predicted results indicated that the top four IMFs based absorption spectra acquired higher R (0.996 1) and lowered RMSEP (0.019 8) compared tothe single PLS regression and theother top several IMFs’results. Thus, the successful application with EMD-PLS method manifests the effectiveness in quantitative analysis of binary mixtures within the THz region.