Abstract: The fast reactor plays a dominant role among the candidate reactor types for the fourth generation of reactors. Developing fast reactor core physics design methods with good computational accuracy is of great significance for the physical design of fast reactors. Among these methods, the calculation method of fast reactor few-group cross-sections is one of the core elements affecting the computational accuracy of fast reactors and is one of the important research directions in the field of reactor physics. To further improve the applicability and computational accuracy of traditional fast reactor few-group cross-section calculation methods, research was carried on few-group cross-section calculation methods based on the fast reactor core physics calculation program MOSASAUR. On the basis of the original zero-dimensional and one-dimensional models, a two-dimensional calculation method was investigated with the subgroup resonance method and MOC (method of characteristic) transport method. A geometric pre-processing module that can perform fine modeling and meshing of components was developed. On the basis of the original library, a multi-group subgroup parameter library suitable for the subgroup fitting method was created. The subgroup method was used for resonance self-shielding treatment and the Bondarenko iteration method was used to handle resonance interference effects. The MOC accelerated by GPU was used for transport calculations. In terms of numerical results, verification and confirmation were conducted at both the lattice and core levels. For lattice problems, three typical fuel lattices were used and two complex fuel lattices were designed to compare different few-group calculation methods. The deviations of eigenvalues, pin powers and few-group macroscopic cross-sections were analyzed in comparison with the Monte Carlo program. The numerical results show that the newly developed two-dimensional few-group cross-section calculation method has good computational accuracy. In particular, for lattices with complex structures, the eigenvalue deviation can be reduced from 10 000 pcm to 11 and 34 pcm, demonstrating good computational accuracy. For the core problem, the MET-1000 and MOX-1000 benchmarks are used for verification. Few-group macroscopic cross-sections required for MOSASAUR core calculations were generated using both the Monte Carlo program and the MOSASAUR program for comparative verification. The numerical results show that there is an error accumulation phenomenon during the burnup process. However, the maximum eigenvalue deviation in the entire initial period is 160 pcm, which demonstrates that the few-group cross-section generation method has good computational accuracy. This method can provide more accurate few-group calculation parameters for core calculations, thereby enhancing the computational accuracy of fast reactor core physics design and offering a new option for fast reactor few-group cross-section calculations.