It could effectively reduce sample redundancy and improve reconstruction accuracy by selecting representative colors as spectral reconstruction training samples. Representative colors were selected as training samples in the Munsell color set by orthogonal experimental method based on color properties, including hue, value and saturation, and the effects of color properties on reconstruction accuracy were analyzed. The results showed that the mean of reflectance RMS could be up to 0.120 4 by PCA method, while it was relatively smaller by PSE method and R-matrix method. Value priority level was greater than the hue and saturation by range analysis. In conclusion, the effects of color properties on PCA method were greater than PSE method and R-matrix method, and value exerted more effects on reconstruction accuracy than hue and saturation.