Boron neutron capture therapy (BNCT) is expected to be an effective method of improving the treatment results on malignant brain glioma and malignant melanoma, for which no successful treatment has been developed so far. The beam shaping assembly (BSA) of accelerator-based boron neutron capture therapy (A-BNCT) consists of a moderator, a reflector, gamma and thermal neutron shielding and a collimator. The BSA moderates the fast neutron produced in target to epithermal energy range. Design of BSA is one of the key jobs in BNCT project. An optimized study was conducted to design a beam shaping assembly for BNCT facility based on 3.5 MeV 10 mA radio-frequency quadrupole proton accelerator at Dongguan Neutron Science Center. In this simulation work, the neutron produced from the 7Li (p, n) 7Be reaction by 3.5 MeV proton is adopted as a neutron source term. In order to search for an optimized beam shaping assembly for accelerator-based BNCT, Monte Carlo simulation is carried out based on the parameters of moderator material and structure, the Gamma shielding, and the thermal neutron filter in the beam shaping assembly. The beam shaping assembly in this work consists of various moderator materials, teflon as reflector, Bi as gamma shielding, 6Li as thermal neutron filter, and lithium polyethylene as collimator. After comparing the simulation results of Fluental and LiF moderator materials, the beam shaping assembly configuration based on sandwich Fluental-LiF configuration is proposed. The sandwich Fluental-LiF configuration is made up of Fluental and LiF layer by layer, like a sandwich structure, and each layer is 2 cm thick. According to the beam quality requirement of the IAEA-tecdoc-1223 report, the optimized epithermal neutron flux in air at the exit of BSA of the sandwich Fluental-LiF configuration is 9.14×108 n/(cm2·s), which is greater than those of the Fluental configuration (7.81×108 n/(cm2·s)) and LiF configuration (8.79×108 n/(cm2·s)), when the ratio of fast neutron component to gamma ray component to thermal neutron is less than the limiting value of IAEA recommendation. Subsequently, the depth distribution of the equivalent doses in the Snyder head phantom is calculated to evaluate the treatment characteristic. The simulation results show that the therapy rate of the beam shaping assembly based on the sandwich Fluental-LiF configuration is basically equal to that of the Fluental configuration and better than that of the LiF configuration, and the therapy time is less than that of the Fluental configuration. This means that the beam shaping assembly based on the sandwich Fluental-LiF configuration is one of the suitable options for our accelerator-based BNCT.