The Monte Carlo method has been widely used in nuclear reactor physics. However, the traditional Monte Carlo method suffers from slow convergence and statistical uncertainty, usually leading to a long computation time. The fission matrix acceleration method has been proposed to accelerate the traditional Monte Carlo criticality calculations, however, it still requires significant computational resources to build fission matrices. A new approach, used in the RAPID code, utilizes a fission matrix combination method to estimate the true fission matrix by combining a set of pre-calculated fission matrix coefficients. Except for the Monte Carlo calculation of the pre-calculated fission matrices, the combination does not rely on the Monte Carlo simulation and saves a large amount of computational costs for subsequent calculations. However, the fission matrix combination method can have significant errors in estimating the source distribution especially with a large enrichment gap between fuels. In this paper, we will explain the issues of the fission matrix combination method in source distribution calculations and present a new method to correct the combined fission matrices. The method is tested on a model with two different enrichment assemblies to calculate both the multiplication factor and the source distribution. The source distribution derived from the corrected fission matrix shows a good agreement compared to the reference, which is the Serpent calculation.