In this study, color similarity metrics in a spectral space are considered. The paper gives a brief overview of several existing measures and presents a novel approach based on kernel methods. New similarity measures for spectral images based upon kernel methods include polynomial, Gaussian radial basis function and sigmoid kernels. The performance of the methods is tested against the Munsell Matte spectral dataset. Kernel methods are compared to twelve well-known similarity metrics, i.e. Correlation Coefficient, Exponential Similarity, Maximum-Minimum methods, etc. Spectral differences of constant Hue, adjacent Values and Chromas have been evaluated using these metrics. The tests show that the proposed Gaussian radial basis function kernel metric performs significantly better, compared to the rest of the measures.