Spectral characterization technique has a prominent advantage that it does not suffer from the problem of metamerism in comparison with Colorimetric characterization methods. PCA (Principle Component Analysis) is an important and useful mathematical method for data reduction, in which a set of spectra, so-called statistical colorants, can be derived from spectral properties of a large set of samples. The spectral reflectance of the color, an admixture of these statistical colorants, can be represented by approximately linear addition of their spectral reflectances. In this paper, a new method for spectral characterization of a flat panel color scanner using PCA method was proposed. Firstly, the PCA algorithm was applied to estimate the spectral reflectance of the statistical colorants on the color targets scanned, and then the colorant scalars were calculated. Secondly, the relationship between the colorant scalars and the scanner RGB signals was built using BP (Back Propagation) neural network. The scanner was characterized also using polynomial regression model and BP neural network directly between scanner RGB values and divice-independent tristimulus values. The experiment results showed that the spectral characterization using PCA method was more accurate than the polynomial regression model and similarly accurate as the direct neural network method but more useful because of the accurate spectral reflectance estimation ability.