To face this challenge, we have developed a wide data base, including enough signatures from several anonymous signers and skilled forgeries. The collected database is multimodal as it also includes fingerprints and voices from all users. We have carried out a State of The Art in biometric recognition at different levels, starting from a wide vision of biometric recognition, evolving through the handwriting recognition and ending in both ways of handwritten signatures. Our proposal for on-line signature recognition is a system based on processing local features (coordinates, velocities, accelerations, pressure and pen tilt) by using global features (center of mass of ink and principal inertia axes) as a reference system. We propose a simple scaling algorithm for signature time normalization that is applicable to every local feature. We have limited the amount of local features to feed the Hidden Markov Models to nine, in order to balance the required security level and the processing and storing capacities. Using six stated LR-HMMs we have obtained equal error rates similar to those observed in the state-of-the-art. We have studied the reference system that is most suitable for the alignment of signatures and we have proposed the reference system located in the center of mass of ink using the principal inertia axes as orientation, after concluding that using signatures first points coordinates and slope as an alignment reference becomes a source of noise. The new reference system provides a better alignment, improving the equal error rate significantly, especially when using only coordinates. Our proposals for off-line signature recognition consist in two systems based on using the LR-HMM technique, which is so useful for dynamic systems. The models are fed on spatially ordered point sequences and geometrical “false-dynamic” derivative features. The goal of those proposals is to extend LR-HMMs to the field of static or off-line signature processing using results provided by image connectivity analysis, which separates images in connected components known as “blobs”, each one made up of a cluster of adjacent pixels of the same nature. Two different ways of generating models are discussed, depending on the way that blobs provided by the connectivity analysis are ordered. In the first proposed method, blobs are ordered according to their perimeter length. In the second proposal, the sorting criterion is based in the natural reading order. Results obtained in different experiments are presented, showing that the proposed methods provide verification rates similar to those observed during the study of the state-of-the-art. The models based on the second criterion are especially adequate for the Latin writing in which signatures have been written.