In this paper, the preprocessing stage of a surface mounted device (SMD) image classification system (ICS) is presented for improvement. The ICS uses two images of the same scene taken using two different types of illuminations, top and side. For each scene, we encounter one of three different cases: SMD present, SMD absent with a speck of glue, and SMD absent with no speck of glue. After areas of improvement to the ICS are identified, a methodology is presented to define and evaluate preprocessing methods. The methodology first defines criteria to improve the images for a fixed case and from it two new methods, the NewC and NewB, are proposed for preprocessing. The existing system, which uses a simple subtraction operation to combine the images, is also evaluated since it serves as a reference. The NewC method applies edge enhancement techniques to the available images before subtraction and the NewB method uses only the side illuminated image using also edge enhancement. The methodology then requires the development of image measurement descriptors that are computed for each preprocessing output image for the SMD present cases in the training database. Some descriptors use the Radon Transform to describe SMD edges in the images and others use energy or a signal-to-noise ratio measure. From the descriptors, image improvement indicators are developed and computed. These are statistical measures of data dispersion applied to the distributions of one or more descriptors and which allow us to assess preprocessing systems. Both NewC and NewB methods are clearly shown to be superior to the ImageC method. The NewC and NewB methods are clearly shown to be superior to the ImageC method. The NewC method is slightly better than the NewB method indicating that very little use is made of the information in the top illuminated image.