Spinal pedicle screw placement is a common method to treat various spinal diseases, such as traumas, tumors, deformities and degenerative illness, but improper screw placement can lead to serious complications. The accuracy of pedicle screw fixation assisted by a robotic system is better than that of manual procedures under fluoroscopy guidance. However, to enable an image-guided robotic surgery we need to plan screw path at pre-surgical. In this paper, we propose a multi-directional reconstruction approach for screw path planning. An optimal screw path in a projected pedicle image is estimated first. Then, the 3D path is a reconstruction from the paths in the projections from multiple directions. Our test results show that the mean angle deviation is (mean ± standard deviation) 4.48±0.21°, the maximum point-to-point deviation distance is 2.40±0.10mm which is located at P₁ and the mean value of the minimum deviation distance is 0.96±0.05mm which is located at P₁₂. The mean value of the maximum tolerance difference is -0.0022mm±0.04mm, which indicates that the value of calculated trajectories is slightly better than the value of the trajectories by the doctors. These results were also confirmed to be within acceptable range by the doctors.