We propose a simple but useful method to represent multispectral images captured by multispectral cameras (MSC) with different numbers of bands. When considering accurate color reproduction under an arbitrary illuminant, it is necessary to represent them in a common space with sufficient accuracy of spectral information, estimated from each multispectral image. The principal component analysis (PCA) is useful to reduce the high dimension of the spectral information. However, PCA-based methods may cause both large differences in dynamic range between coefficients of the basis functions derived from the PCA and negative pixel values, therefore, it is not easy to handle or edit coefficient images. To solve the problem, we propose the idea of a virtual multispectral camera (VMSC) that transforms real multispectral images into virtual multispectral images. We design the sensitivities of the VMSC properly, and our unified representation can avoid some disadvantages of conventional PCA-based methods. We experimentally demonstrate the color reproduction accuracy of our method by comparing it with the PCA-based methods, and we show an example of a virtual multispectral image transformed by our method.