Abstract: In order to accurately extract the navigating path of virtual endoscopy and eliminate the rotation of visual images in the navigating process, a method for virtual endoscopy automatic navigation that combines skeleton extraction and path optimization is presented in this paper. Firstly, a shortest path generation algorithm based on weighted voxel density is adopted to calculate the simple navigating path from the skeleton of the 3D medical imaging data. Then, the final navigation path is generated by fitting cubic B-Spline curve. In the navigating process, the up vector of the virtual camera is optimized by calculating the minimal rotation frames for navigating path curve. In order to limit rotation angle variation of the up vector, we re-sample the view points on the navigating path by a given threshold dynamically. Experimental results show that the new method can extract accurate navigating path, and give good navigating visual effect.