Bone structures stand out from the surrounding areas in terms of intensity, which facilitates their segmentation. On the other hand the segmentation of a single bone structure can be very challenging, because other bone structures are usually very close and in case of the vertebrae they have a very similar shape. Therefore a priori knowledge about the structure positions has to be included for a proper segmentation. For this purpose Fraunhofer IGD has developed an “Articulated Atlas” model. From annotated training data it learns the shape and texture of single bones and the relative positions of the bones to each other. This knowledge is then used during the adaptation phase on the data set, which is going to be segmented, to limit the segmentation to plausible bone constellations. After successful segmentation, the shapes and locations of the single vertebrae are known and outlined.