Objective This study was to investigate the mechanism of action of polycaprolactone/gelatin (PCL/GE) composite fiber scaffold with nerve growth factor (NGF) in the recovery of spinal cord injury (SCI). Methods Sixty female Sprague-Dawley (SD) rats were randomly assigned to the negative control group, the positive control group, the PCL/GE scaffold group, and the collagen-binding structural domain nerve growth factor (CBD-NGF)/PCL/GE scaffold group, with 15 rats in each group. Spinal cord transection was used to establish SCI models in rats. The negative control group received sham surgery, while the other three groups were given spinal cord transection at the tenth thoracic vertebra (T10) segment. The rats in the PCL/GE scaffold group were implanted with a 4 mm PCL/GE composite fiber scaffold, and those in the CBD-NGF/PCL/GE scaffold group were implanted with a CBD-NGF/PCL/GE composite fiber scaffold. The Basso–Beattie–Bresnahan (BBB) locomotor rating scale was used to evaluate the locomotor ability of the hind limbs of the rats, and the amplitude and latency of motor evoked potentials (MEP) were recorded by neurophysiological testing at 12 w postoperatively. The levels of growth-associated protein 43 (GAP43) and neurofilament protein 200 (NF200) in the spinal cord tissue of the injury site were determined using Western Blot at 12 w after surgery. Spinal cord tissues of 2 cm within the injury site, the thoracic segment above the injury site, and the lumbar segment below the injury site were collected from the measurement of axonal transport using fluorescent retrograde tracer fluorogold, and the integrated absorbance (IA) values of FC-positive cells were calculated. Results After treatment, the negative control rats showed normal locomotion function of the hind limb with the highest BBB scores, while the positive control rats had the lowest BBB scores and showed paraplegia. The scaffold groups exhibited better locomotion function of the hind limb and higher BBB scores than the positive controls, with greater improvement observed in the CBD-NGF/PCL/GE scaffold group (P < 0.05). Compared with the positive controls, the PCL/GE scaffold group and CBD-NGF/PCL/GE scaffold group exhibited significantly shorter latency and increased amplitude of MEP, with more significant changes observed in the CBD-NGF/PCL/GE scaffold group (P < 0.05). Compared with the positive control group, the GAP43 and NF200 levels of spinal cord tissue were significantly elevated in both the PCL/GE scaffold group and the CBD-NGF/PCL/GE scaffold group, and the changes were more pronounced in the CBD-NGF/PCL/GE scaffold group (P < 0.05). The differences in the IA values of FC-positive cells in the spinal cord tissue of the lumbar segment below the injury site among the four groups did not come up to the statistical standard (P > 0.05). Compared with the positive control group, the FC-positive cell IA values of spinal cord tissue in the thoracic segment above the injury area were markedly increased in the PCL/GE scaffold group and the CBD-NGF/PCL/GE scaffold group, and the alterations were more significant in the CBD-NGF/PCL/GE scaffold group (P < 0.05). Conclusion PCL/GE composite fiber scaffold with NGF significantly improves motor and neurological functions in the hind limbs of SCI rats and promotes the recovery of axonal transport, and the mechanism may be associated with the upregulation of GAP43 and NF200 levels in spinal cord injury site tissues.