In this paper, a novel system for off-line signature verification is proposed. We introduced a complete set of preprocessing algorithms. The proposed alignment, orientation and thinning algorithms have proved to give excellent and fast preprocessing of the signature. Due to the lack of an international signature database (for privacy reasons), we have to make our own database. The database has genuine signatures, casual (common) and skilled forged signatures for each individual. The feature extraction and verification of processed signatures are implemented using principle component analysis (PCA) and back propagation neural networks respectively. The system is tested using both casual and skilled forgeries. For a threshold of 90%, the network gives a zero false rejection ratio (FRR) and a false acceptance ratio (FAR) of about 1% for skilled forged signatures