The 3D reconstruction of the human spine is essential for accurate diagnosis of deformities like scoliosis. Here, a low cost stereo-radiographic 3D reconstruction method is proposed. A generic model is built using biomedical modeling of a cadaveric spine. Biplanar radiographs are obtained from the subject using a calibration apparatus. Six stereo-corresponding and nineteen non stereo-corresponding points per vertebra are identified and their 3D positions are computed. Using these landmarks, the generic model is deformed to obtain a personalized spine model using dual kriging algorithm. A novel technique is proposed for evaluating spinal deformities like axial vertebral rotation and spinal curvature. First, the axial vertebral rotation is found using vertebra vector parametric calculations. Proposed semi-automated method needs identification of only two landmarks per vertebra. The most tilted vertebra is used to find the plane in which spinal curvature is maximized. At this plane, the spinal curvature is automatically computed using a projection technique. The 3D spine reconstruction thus obtained is validated by back projecting it on both the x-rays. A close match among the projected and real vertebral contours are observed. Also, a quantitative validation of 3D reconstruction and deformity quantification techniques have been presented. Thus, a low cost 3D diagnostic procedure is developed for spinal deformity quantification