Abstract: In the fusion reactor, the first wall, blankets and superconducting magnets are directly exposed to 14.5 MeV neutrons produced by fusion reaction. Some of the characteristics of Nb metal such as high melting point, good thermal conductivity and low neutron capture cross section make Nb often used in these parts of reactor. Therefore, the quality of the evaluated nuclear data of Nb is required more accurate to ensure the safety and economy in the process of the fusion reactor working. Benchmark experiments are an important way to test and evaluate the reliability of data. A benchmark experiment on Nb was carried out with 14.5 MeV DT neutron source in China Institute of Atomic Energy. Slab Nb with thickness of 5 cm, 10 cm and 15 cm was selected as samples. For each thickness of Nb, the leakage neutrons from 0.8 MeV to 16 MeV were measured using the timeofflight method in the 60° and 120° directions. The Monte Carlo neutron transport code MCNP4C was used to simulate the leakage neutron spectra under the same experimental conditions, the Nb data was retrieved from CENDL-3.1, ENDF/B-Ⅷ.0 and JENDL-4.0 evaluation data libraries. The whole time of flight spectrum was divided into the parts contributed by four reaction channels including (n, el), (n, inl)D, (n, inl)C, and (n, 2n). The simulation integral value of each part was compared with the experimental integral value, and the C/E was obtained as the standard to verify the reliability and accuracy of the relevant data. As the results of the benchmark experiment on Nb, it can be found that: Except a little overpredict at the simulation result of elastic scattering at 120°, the Nb nuclear data of the JENDL4.0 data library is in good agreement with experiment results. One of the problems of CENDL3.1 library are that it gives too large discrete inelastic scattering cross section. The other problem may be that the secondary neutron energy spectrum given by continuous level of inelastic scattering and (n, 2n) reaction is soft. The simulation results of the ENDF/B-Ⅷ.0 data library make a huge difference. It can be seen from the comparison results that the cross section of the energy spectrum in the discrete inelastic scattering part is high and the energy spectrum given in the continuum inelastic scattering part is also problematic. The discrepancies of elastic scattering peak between 60° and 120° are considered as caused by the wrong angular distribution, especially the underestimate at 60°. So the ENDF/BⅧ.0 library has a lot to improve. In a conclusion, these three libraries are all need to be improved to get more accurate data.