A novel technique is presented for off-line signature recognition and verification. The feature extraction procedure employs directional-vectors, similar to those used in chain codes, which provide a global measure of the signature image. The signature trace is transformed into the feature vector by measuring the directional strength of line segments having a chessboard distance equal to two. A probabilistic neural topology is employed for the design of the classifier. In order to obtain comparable results, the method was applied to a database already used in the literature. The verification procedure provides low classification error for authentic signatures while it eliminates the forgers.