In the above-described first embodiment, the sample color for which the spectral reflectances of the characteristics-acquired gradation value for the prediction target color can be predicted with highest accuracy using the first relational equation based on the characteristics thereof is selected, as the reference color, from among the plurality of sample colors, and the spectral reflectances of the prediction target gradation value for the prediction target color are predicted using the second relational equation based on the characteristics of the reference color. With such a configuration, some error occurs between the prediction values and the actual measurement values regarding the characteristics-acquired gradation value. For example, in a case in which eleven gradation values in increments of 0.1 from the gradation value of 0 to the gradation value of 1 are the prediction target gradation values, and only the gradation value of 0.5 is the characteristics-acquired gradation value, the spectral reflectances of the eleven gradation values (prediction target gradation values) are predicted using the second relational equations based on the characteristics of the reference color. In this case, some error occurs between the prediction values and the actual measurement values regarding the gradation value of 0.5. Herein, if the actual measurement values are adopted for the gradation value of 0.5 in order to set such an error for the gradation value of 0.5 to 0, then as seen from