The work aims to study the spectral information of the original image during the spectrum color reproduction and reconstruct the spectral reflectance of the target color, so as to explore the factors affecting the accuracy of reconstructed spectra. With two kinds of color cards (Munsell Color Matt (1269 color lump) and Color Checker Classic (24 color lump)) selected as the spectral reflectance data samples, different linear reconstruction models of PCA were established and different numbers of base vectors were selected to separately reconstruct the spectra. Then, their accuracy was evaluated. Classic color card was taken to simulate the target color in the multispectral image that reconstructed the spectral reflectance, so as to study and compare the influence of the number of spectral reconstruction models and base vectors on the reconstruction accuracy. The experimental results showed that the finally restored data by dimension reduction model 1 were better than dimension reduction model 2 in RMSE, GFC and color difference. With the increase of the number of base vectors, the gap of two dimension reduction models was reducing. When the number of base vectors was more than 13, the two models were almost no difference. The proposed spectral reconstruction model 1 and 7 base vectors are the optimal scheme for reconstructing spectral images.