The intermingling of regenerated nerve fibers inside nerve grafts is the main reason for mismatched nerve fibers. This is one of the key factors affecting limb function recovery after nerve injury. Previous research has shown that the accuracy of axon regeneration can be im-proved by a bionic structural implant. To this aim, iodine and freeze-drying high-resolution micro-computed tomography was performed to visualize the 3D topography of the New Zealand rabbit sciatic nerve (25 mm). A series of 1-, 2-, 3-, and 4-custom anatomy-based nerve conduits (CANCs) were fabricated based on the anatomical structure of the nerve fascicle. The match index, luminal surface, and mechanical properties of CANCs were evaluated before implanting in a 10-mm gap of the sciatic nerve. Recovery was evaluated by histo-morphometric analyses, electrophysiological study, gastrocnemius muscle weight recovery ratio, and behavioral assessments at 12 and 24 weeks postoperatively. The accuracy of nerve regeneration was determined by changes in fluorescence-labeled profile number during simultaneous retrograde tracing. Our results showed that the optimal preprocessing condition for high-resolution micro-computed to-mography visualization was treatment of the sciatic nerve with 40% Lugol’s solution for 3 days followed by lyophilization for 2 days. In vitro experiments demonstrated that the match index was highest in the 3-CANC group, followed by the 2-, 1-, and 4-CANC groups. The lumi-nal surface was lowest in the 1-CANC group. Mechanical properties (transverse compressive and bending properties) were higher in the 3-and 4-CANC groups than in the 1-CANC group. In vivo experiments demonstrated that the recovery (morphology of regenerated fibers, compound muscle action potential, gastrocnemius muscle weight recovery ratio, pain-related autotomy behaviors, and range of motion) in the 3-CANC group was superior to the other CANC groups, and achieved the same therapeutic effect as the autograft. The simultaneous retrograde tracing results showed that the percentages of double-labeled profiles of the 2-, 3-, and 4-CANC groups were comparatively lower than that of the 1-CANC group, which indicates that regenerated nerve fascicles were less intermingled in the 2-, 3-, and 4-CANC groups. These findings demonstrate that the visualization of the rabbit sciatic nerve can be achieved by iodine and freeze-drying high-res-olution micro-computed tomography, and that this method can be used to design CANCs with different channels that are based on the anatomical structure of the nerve. Compared with the 1-CANC, 3-CANC had a higher match index and luminal surface, and improved the accuracy of nerve regeneration by limiting the intermingling of the regenerated fascicles. All procedures were approved by the Animal Care and Use Committee, Xinjiang Medical University, China on April 4, 2017 (ethics approval No. IACUC20170315-02).