A pattern recognition based technique has been used to classify the different constituents (macerals) of coal. The method has two modules, an off-line training module and an on-line classification module. Three-band color (R-G-B) images of coal samples are used in contrast to monochrome (B and W) images used by earlier researchers. Each maceral class is determined based on the gray values (or the reflectance properties) in the three bands. Points are assigned to one of vitrinite, exinite, inertinite or background class using a minimum distance classifier. This means that the distance of a point from its assigned cluster centre is minimum with respect to the distance of the point from other cluster centres. This method also minimizes the sum of squared error of mis-classification. The method has a quantitative basis to standardize petrological analysis of coal samples in relation to reflectance properties of different macerals under a given lighting condition. The speed of the system is considerably faster than the current technique of analysis and it ensures repeatability of results.