ABSTRACT We propose an approach using a radial segmentation method and Kohonen's SOM (self-organizing map) to resolve the off-line Chinese-based signature verification problem. The radial segmentation method extracts features of a signature in a consistent manner regardless pen thickness, size and rotation. The threshold SOM trains only on genuine signatures of an individual. Verification is achieved by determining if a test signature is within the threshold boundary for a cluster generated. An experiment including 10 subjects and 100 forgers was conducted for testing this approach. As a result, an average false rejection rate (FRR) of 8% and a false acceptance rate (FAR) below 3% are achieved, depending on the number of hidden nodes involved in training. The approach demonstrates the flexibility in choosing error rates of interest and capability in achieving low error rates.