In order to accurately and efficiently extract skeletons from 3D images, a new parallel 3D skeletonization algorithm is proposed. The distances of original images are firstly transformed, and then iterative parallel thinning is performed. Each iteration is divided into 6 sub iterations, one sub iteration is used to process a kind of boundary points,and the foreground points satisfying the primary selection rule are labeled. After 6 sub iterations are completed, labeled points are rechecked in an ascending distance order,and those who satisfy thinning conditions are deleted. This process is repeated until there is no point deleted. This algorithm is applied to 2D and 3D images, obtained skeletons are consistent with that by the visual apperception. This algorithm guarantees the connectivity and topology of extracted skeletons, and its median can be validated by caculating the distance values of skeleton points,which reflects the essence characteristics of the object.