Natural food pigments are widely used, and trichromatic pigments can be mixed with different colors according to their concentrations. Traditional color matching methods are highly dependent on experience of color matching, which result in low production efficiency, big variation and poor product quality stability. In this study, according to the characteristic absorption peak of natural food pigments had no change during mixing, and based on the matching method of absorption spectrum, color matching models of multiple linear and neural network were established between the concentration of pigments and the absorption value of the characteristic absorption peak. And the optimal model was selected and tested by errors analysis. The results showed that the prediction accuracy and stability of neural network model were better than those in multivariate linear model, and the color differences between the prediction formula and the original formula were within 3, which could not be distinguished by naked eyes. Thus the neural network models were more suitable for color matching requirements. These results provided theoretical basis for the intelligent color matching of natural food pigments.