Abstract : The problem of Signature recognition which is a very special case of image analysis describes how to design a system that extracts the features efficiently and verified the signature perfectly. Thus, most of our attention is devoted to develop preprocessing the signature's image, features extraction techniques, and the classification method. In this work, a robust technique for the off-line signature verification problem in practical application is presented. It possesses many important advantages especially in image processing and feature extraction. In image processing, many technologies are used, through which the signature is extracted from the original image, filtered from any noises, image improvements that prepare the signature to the feature extraction stage, and ending with vector of values that describes the signature image features. Then, these obtained values are included in the Learning Vector Quantization Neural Network (LVQ-NN) which, in turn, will characterize the signature to whom it belongs. The system recognizes signatures correctly without being affected by shift in position, and scaling. The system is also recognizes signatures with changes in angle and expression. Small set of images per person in the training database is needed to produce acceptable classification accuracy. The results obtained show that the proposed system gives a very encouraging performance. During the work, Special image techniques such as noise reduction, size normalization, and signature thinning are proved. These effects are investigated and treated in this thesis to make the system unaffected by noise, signature's scale, and size variations. The proposed system structure was implemented as a software using MATLAB version 7.