Abstract: Under irradiation of high-energy particles (neutrons or ions), cascade collisions continue to occur in nuclear materials, producing a large number of vacancy clusters and interstitial clusters. These defect clusters will undergo reactions such as diffusion, aggregation, recombination, and annihilation, forming important microstructures such as voids and dislocation loops. Cluster dynamics (CD) method is one of the important methods for simulating the microstructure evolution of nuclear materials under high-energy particle irradiation. The source term (i.e., the average size distribution of huge number of cascade collision defect clusters) is key input of the CD method. In the classical CD method, the source term is usually obtained by empirical fitting, and the maximum cluster in the source term usually only contains several vacancies or interstitials. According to recent studies on cascade collision simulations based on molecular dynamics (MD) methods, cascade collisions corresponding to high-energy primary knocked atom (PKA) can form clusters with dozens or even hundreds of vacancies or interstitials. So, the initial defect information obtained at the atomic level was not fully utilized in traditional CD simulations. With the development of molecular dynamics, cascade collision defect databases have been greatly enriched. Combined with PKA energy spectrum, it is enough to obtain more reasonable source term. The PKA spectrum is generally a quasi-continuous spectrum, and the number of energy groups is usually tens or hundreds. However, the MD cascade collision defect database usually only contains about ten energy values, which is much smaller than the former. If the MD cascade collision defect database is to be used rationally, a mapping rule from a quasi-continuous spectrum to a finite number of discrete energy values must be determined. In view of this, based on probability theory and the principle of proximity, fully considering the changing characteristics of the PKA spectrum curve, five sampling algorithms were proposed for obtaining discrete values of cascade energy from the quasi-continuous PKA energy spectrum. Based on the probability of discrete values of cascade energy, the defect cluster size distribution in the cascade collision defect database was weighted, grouped, fitted, and a relatively smooth continuous size distribution curve was obtained as the source term of the CD method. In order to test the rationality of the source term algorithms, the CD method was used to simulate the low-dose neutron irradiation of pure tungsten experiment in the HFIR reactor, and compared the defect cluster information. The results are consistent with the experimental data. Since there are differences in the size distribution of defect clusters after irradiation simulation with the five algorithms, after analysis, it is believed that the fifth source term algorithm is the most reasonable. In addition, the shortcomings of the current source term algorithm and CD model used and improvement that can be expected in the future were also summarized.