We have developed a thoracoscopic surgical navigation system for lung cancer localization. In our system, the thoracic cage and mediastinum are localized using rigid registration between the intraoperatively digitized surface points and the preoperative CT surface model, and then the lung deformation field is estimated using nonrigid registration between the registered and digitized point datasets on the collapsed lung surface and the preoperative CT lung surface model to predict cancer locations. In this paper, improved methods on key components of the system are investigated to realize clinically acceptable usability and accuracy. Firstly, we implement a non-contact surface digitizer under thoracoscopic control using an optically tracked laser pointer. Secondly, we establish a rigid registration protocol which minimizes the influence of the deformation in different patient's positions by analyzing MR images of volunteers. These techniques were evaluated by in vitro and clinical experiments.