Virtual endoscopy, which is a noninvasive procedure for detecting anomalies inside human organs, is meaningful for medical diagnosis and surgery. In order to perform an accurate navigation, the centerline of the model must be extracted. In this paper, a new centerline extraction algorithm based on Hessian Matrix is proposed. First, the distance transformation is performed. Then the initial path is obtained by computing the eigenvalues and eigenvectors of the Hessian matrix. After that, the visibility test with an adaptive visibility sphere radius, which is determined by the eigenvalues of the Hessian matrix, is carried out to remove the useless voxels in the centerline. Finally, the path with all the points staying away from the surface is generated by Dijkstra抯 shortest path algorithm. The experimental results illustrate the efficiency of the proposed algorithm.