In order to address the automatic virtual reassembling of 3D thin-wall fragments, this paper proposes a 3D fragment reassembly method based on derivative dynamic time warping. Firstly, a calculation method of discrete curvature and torsion is designed to solve the difficulty of calculating curvature and torsion of discrete data points and eliminate effectively the noise interferences in the calculation process. Then, it takes curvature and torsion as the feature descriptors of the curve, searches the candidate matching line segments by the derivative dynamic time warping (DDTW) method with the feature descriptors, and records the positions of the starting and ending points of each candidate matching segment. After that, it designs a voting mechanism with the geometric invariant as the constraint information to select further the optimal matching line segments. Finally, it adopts the least squares method to estimate the rotation and transformation matrices and uses the iterative closest point (ICP) method to complete the reassembly of fragments. The experimental results show that the reassembly error is less than 1mm and that the reassembly effect is good. The method can solve the 3D curve matching in case there are partial feature defects, and can achieve the virtual restoration of the broken thin-wall fragment model quickly and effectively.