Many old Asian paintings have been drawn with natural mineral pigments. The digital archive of those paintings, the identification of pigments used in the paintings and the retrieval of color fading are strongly desired. The multispectral image acquisition of those paintings has a great potential in the analysis of the used pigments. This paper describes the segmentation of multispectral images of those paintings. Here we propose to use the kernel subspace method for this purpose. First the dependency of spectral reflectance of natural mineral pigments on grain size is shown and then segmentation results for the images of color patches and a Buddha painting are demonstrated.