The spectral reflectance of the 24, 140, 1269 matt Munsell color cards were analyzed respectively by principal component analysis, and then spectral reconstruction was carried out using 6 biggest eigenvectors extracted by the analysis. The results were evaluated by using RMSE and color difference value of CIE1976. The reconstruction results showed that 140 color cards surpass other two kinds of quantities in accumulative total contribution rate and color difference under the same reconstruction condition; RMSE of the 140 color cards is inferior to that of the 1269 color cards, which suggests that the number of training sample has influence on reconstruction precision; reconstructed spectral reflectance is in the middle band of yellow, green, and blue, which represent the best reproduction; the reproduction performance of blue-purple is not as good and that of orange-red performance is the worst.