Despite recent advances in the areas of fingerprint identification, fingerprint matching continues to be a chanllenging pattern recognition problem. The first step to this problem is the extraction of landmarks known as minutiae points from a print. Once extracted, these points are then compared to all sets on file in search of a match. The accurate extraction of minutiae from an image is the basis for the entire matching process. Various minutiae extraction approaches have been proposed in the literature, each with its own merits and degree of sucess. The most common approach is to extract the ridges in the fingerprint image through skeletonization, apply ridge following, and use rule-based classification for minutiae detection. Our emphasis in this paper is on extracting the minutiae from the original gray-scale images, without any image preprocessing. In particular, we have implemented and compared three methods based on elgenspace representations and neural network classifiers. Moreover, we present preliminary results of an attempt to fuse the outputs of these three methods using a clustering algorithm unique to this type of problem.