Architecture recovery is crucial to supporting software maintenance and evolution. The clustering problem that could implement architecture recovery is considered as optimizing problem in this paper. Through improving important parameters and core steps of general genetic algorithm, such as initial population, select operator, self-adapting ability of crossover probability and mutation probability, a hybrid genetic clustering algorithm (HGCA) is designed and implemented. An experiment is given to analyze the availability, effectiveness and synthetical performance of the algorithm. The results show that compared to general GA, the HGCA can produce good initial population, better convergence efficiency and convergence precision. Moreover, the value of the MoJo similarity metrics presents the correctness and effectiveness of HGCA recovering software architecture.