Data-driven approach is an appealing way to depict people in a virtual world. The captured shape and movement data from real people are structured and combined to reproduce or create new samples in an intuitive and controllable way. We focus on the body shape modeling and elucidate the issues related to data-driven methods. The difficulty of adopting data-driven approach for human body shape modeling is due in part to the intrinsic articulated structure of the body. Since such internal structure is not measured with most of existing capture devices available today, it has to be calculated through estimation. We develop a framework for collecting and managing range scan data that automatically estimates this structure from user-tagged landmarks. By framing the captured and structurally annotated data so that statistic implicit is exploited for synthesizing new body shapes, our technique support time-saving generation of animatable body models with high realism.