Segmentation is the first step when trying to use radiological image data in many clinically important applications as radiological diagnosis, monitoring, radiotherapy and surgical planning. Especially in case of large 3D medical data sets the availability of efficient segmentation methods is a critical issue. Incorporation of prior knowledge about the shapes of the organs can speed up the process and can reduce the amount of the user-interaction. This knowledge can be exploited by statistical evaluation of a training dataset. Before statistical analysis can be started, the individual objects have to be registered in a reasonable way. The spatial distribution of a particular point of a bone can be studied only if that point marks the same part of all bones in the learning database. In other words, a point on a particular anatomical landmark has to correspond to the points on the same anatomical location of the other bones. This correspondence problem is difficult, especially for simple objects, where only a little number of landmark points is available. Solving the correspondence based on solely point-wise anatomical landmarks is often impossible. In these cases, we can still exploit curve- or surface-based shape features of the organs.