A line tracking method of arm vein is proposed based on Bayesian theory, and the method can detect vascular boundary. The method can automatically select initial seed points, avoiding artificial disturbance. In the vessel tracking, the vessel structures are divided into normal, bifurcation and crossing types. For each iteration, the algorithm takes longitudinal and horizontal characteristics of vessels into account. Because blood vessels within a short distance are approximately a straight line, a multi-scale line template is applied on the image filtering to obtain line strength of each pixel. A Gaussian model is used to fit the gray distribution of vessel along the cross section. The most possible vessel structure is determined based on Bayesian maximum posteriori probability criterion. Therefore, the edge points, central points, diameter and direction of local vessel can be acquired. The experiment results show that, compared with traditional threshold segmentation method and repeated line tracking method, the proposed method performs better in identification accuracy, comprehensiveness, and robustness to noise.