In this paper, a new spectral characterization method for displays is developed. Firstly, after singular value decomposition, the k most important basis vectors can be obtained for each of the three channels, so that for any given driving signal, its corresponding spectral radiance on each channel can be the linear combinations of these basis vectors and the datum coordinates can be acquired through orthogonality. Then, a lookup table is compiled. Thus, for any driving single in an individual channel, the corresponding spectral radiance can be linearly expressed by basis vectors and the combination coefficients can be got by the lookup table and the interpretation technique. Finally, for any set of driving signals R, G, and B, the overall spectral radiance can be predicted by the summation of the spectral radiances on the three individual channels. When k=1, the proposed method becomes the traditional piecewise linear interpolation assuming constant chromaticity (PLCC) method. The inverse model for the proposed method is also given. Comparison results in terms of the color difference and spectral root mean square error show that the proposed forward model is better not only than the traditional Gain-Offset-Gamma and PLCC methods but also than the newly developed spectral radiance model based on wavelength partition and spectral radiance piecewise partition model.