Knowledge of vertebra location, shape and orientation is crucial in many medical applications such as orthopedics or interventional procedures. The wide range of shapes, joint alterations and pathological cases encountered in an aging population makes automatic segmentation sometimes challenging. This paper presents a new automated vertebra segmentation method for 3D CT data which tackles these problems. This method has two consecutive main steps: first a new coarse-to-fine method produces a coarse shape of the vertebra, then a Hidden Markov Chain (HMC) segmentation using a specific volume transformation refine the segmentation. No shape prior is used thus allowing most frequent non-standard and pathological cases handling. We experiment this method on a set of standard vertebrae and on non-standard cases as encountered in daily practice. After expert validation, we show that our method is robust to shape and luminance changes, and provides correct segmentation for pathological cases.