Abstract: This paper presents an automatic mesh generation method for large-scale parallel finite element analysis in the cluster system. The method is used to prevent parallel efficiency from decreasing caused by imbalance of different subdomains in parallel computation. It has been applied to the partition of a car frame model. The Greedy algorithm and ANP algorithm are discussed and compared around respective character and performance. And also detailed flowchart for algorithms are given. Numerical examples show that this method can give satisfactory results for different elements and therefore has extensive applicability. Without much modification, this method can be used in the dynamic load balancing resulting from various dynamic parallel computing problems.