Abstract: By combining the inherent parallelism of genetic algorithm with the high-speed parallelism of cluster system, a back analysis method of geomechanical parameters based on a master-slave parallel genetic algorithm is presented. Several improved methods are used: real code is used to shorten the length of individual code and reduce the searching scale; dynamic task allocation is used to avoid idle resources of some processors; loose coupling method is used to couple a master-slave parallel genetic algorithm with FLAC. Besides theoretical analysis, the back analysis program is compiled in C and MPI (Message Passing Interface), and it is also tested by a standard elastic problem. The test results indicate that the back analysis program can achieve high precision and approach linear speedup along with the increase of model scale. Therefore, this back analysis method is applicable and propagable, which can satisfy the timeliness of engineering.