Abstract: Fingerprint recognition is a method for biometric authentication. Fingerprint image consists of interleaving ridges and valleys. Ridge termination and bifurcation, uniformly called minutia, are generally used for fingerprint matching. Automatic fingerprint recognition typically goes through a series of processes, including ridge orientation estimation, segmentation, enhancement, minutiae detection and matching. Ridge orientation is one of the fundamental features of a fingerprint image. And orientation estimation is the basis of fingerprint recognition, since it serves for segmentation, enhancement, minutiae extraction and matching. Most existing orientation estimation methods are based on the characteristic of pixel intensity in a block. In this paper neural network is used to learn the ridge orientation. At the training stage, the correct orientations are fed into the network as positive samples, and the incorrect orientations are fed into the network as negative samples. The trained network has the property of responding to true ridge orientation with a large value and of responding to the false ridge orientation with a small value. When estimating fingerprint ridge orientation, the responded values to each orientation at each image block are used to compute the fingerprint orientation field. The proposed method turns out to be more robust than the existing method.