The work aims to improve the reconstruction accuracy of spectral reflectance with principal component analysis (PCA). Matlab was used to conduct simulation experiments. 3 different color cards were selected as the training samples and the effect of the number of principal components and the sample interval on the reconstruction results was investigated by the method of PCA. When the number of principal components was 4, the contribution rate was over 99%. When the sample interval was 10 nm, the RC24 color card had the best reconstruction effect, its average color difference was 2.37 and the average root-mean-square error was 0.0185. The selection of training samples will affect the accuracy of spectral reconstruction. With the characteristics of small amount of data and high accuracy of reconstruction, RC24 color card can be preferred in the field of color reproduction.