This paper describes a method for identifying object materials on a raw circuit board by means of multi-spectral imaging. First, we develop a spectral camera system for observing tiny objects. Second, an algorithm is proposed for estimating the spectral reflectance functions of object surfaces. We do not use the finite-dimensional linear model, but present a direct method under the narrow band assumption. We distinguish metals and dielectrics based on the difference of reflectance in changing illumination geometries. Third, an algorithm is presented for classifying the objects into several circuit elements by using the estimated spectral-reflectances. Finally, region segmentation results are demonstrated in an experiment using a real circuit board.