A method is described to reproduce reflection properties at the surface of real objects using spectral images, which are often saturated because of the limited dynamic range of image detectors. Reflection components are separated into diffuse and specular reflection components based on the dichromatic model at 5-nm wavelength intervals between 380nm and 780nm for each pixel of the spectral images, and diffuse reflectance parameters are estimated during the separation. To estimate specular reflectance parameters, the Gaussian distribution function of the Torrance- Sparrow reflection model is transformed logarithmically to a linear form, and the unsaturated values of the specular reflection components are subjected to the least squares method. Experimental results with a real object demonstrate the efficiency of the proposed method.