Abstract: For deep-penetration shielding problem, it is a great challenge to obtain reliable results with Monte Carlo (MC) method in reasonable time. Local variance reduction (LVR) method and global variance reduction (GVR) method based on discrete ordinate (S_N) method can decrease tally error of deep-penetration problem in MC calculation. Calculation efficiencies of LVR method and GVR method for the HBR-2 benchmark were compared in this paper. Numerical results show that LVR method and GVR method both obtain satisfied results for the HBR-2 benchmark. GVR method can optimize both tallies of radiation surveillance capsule and ex-core detector at once, therefore it is more convenient and efficient.