In order to improve the accuracy and stability of terahertz spectroscopic quantitative analysis of multicomponent mixtures, Tchebichef image moment combined with partial least squares (PLS) regression modeling method was introduced into terahertz spectral quantitative analysis. The mixture of rubber additive zinc oxide, silica and nitrile-butadiene rubber was used as experimental sample. The zinc oxide in the mixture was quantitatively analyzed. After obtaining the terahertz absorbance data of the sample, 3-D spectra of samples were constructed by self-construction method of 3-D spectra. Tchebichef image moments were used to extract the feature information of the 3-D spectral gray image of the sample. Then a quantitative model was established by PLS to quantitatively analyze zinc oxide. The experimental results show that, the correlation coefficient and root mean square error of the sample prediction set obtained by this method are 0.9993 and 0.0351, respectively. Compared with the absorbance data modeling directly by PLS and support vector regression, the results are more accurate and stable. This method can be used for accurate and stable quantitative analysis of zinc oxide in mixtures. It provides a new idea for terahertz spectroscopic quantitative analysis of multicomponent mixtures.