A feature extraction and matching algorithm is proposed based on chain code to study the arm vein. The skeleton structure of the vein is extracted from the near infrared images of the arm and then divided into several curve segments. Matched curve pairs are calculated based on the relative direction, relative location and shape features of curves, and then the spatial transformation between the matched curve pairs is obtained with the particle swarm optimization algorithm. The matching probability is calculated based on the overlapping ratio of all the transformed vein points. The experiment on a database composed of arm images of 110 subjects from 9 countries shows that the identification rates for rank-1 and rank-10% are 74.5% and 93.6%, respectively, which is superior to the results obtained with algorithms of modified Hausdorff distance and template matching. It indicates that arm veins can be used as a new biometric feature for identity recognition.