Characterizing the fracture surfaces within a single sand particle precisely and describing it quantitatively plays an essential role in understanding the breakage behavior of sands. This paper presents a novel method to obtain the realistic fracture surface from 3D particle fragment reassembly. Firstly, a few sets of 2D slices of a fractured particle were collected by using a self-designed mini-loading apparatus combined with the nano-focus X-ray computed tomography. Then the de-noising and watershed segmentation algorithms were applied to the 2D images before the reconstruction work which aimed to acquire 3D digital models of the broken fragments. After separating the 3D fractured particle into pieces, we used the minimum spanning tree to construct the feature curve networks of the fragments that has a higher curvature compared with other locations and to identify all simple chordless cycles in this feature curve networks. The Hausdorff distance and the modified 4-points congruent set algorithms were adopted to identify potential simple chordless cycles for matching and to match them. To avoid the substantial penetration effects, we used the ray-triangle intersection algorithm to detect whether the matched points are within triangulated volume or not. The matching result shows that the proposed method is capable of reassembling fractured particles and it will facilitate the better prediction of particle fracture in numerical modelling. (C) 2018 Elsevier B.V. All rights reserved.