This study evaluates a registration technique for intraoperative tracking of the spine which consists in matching intraoperative measurements with specific surfaces defined on a 3-D preoperative model. The validation study was undertaken on a cadaveric spine. The accuracy for four approaches combining different tracking systems and 3-D reconstruction techniques were compared. The accuracy resulting from the combination of a radiographic 3-D reconstruction and a magnetic digitizer was 5.9±2.7mm. The corresponding errors on vertebral rotations were 4.4±3.3°, 6.7±5.8° and 5.0±3.8° in frontal, sagittal and transverse planes, respectively. This approach is minimally invasive (only 2 X-rays) and may provide sufficient accuracy for certain clinical applications. With CT scan 3-D reconstruction, the accuracy was increased by about 2mm, but the high radiation exposure associated with CT scan imaging for long spinal segments makes it unfavorable to most clinical use. As for the mechanical arm, considering the small increase in accuracy and its awkwardness, its use during surgery is not suggested.