Abstract: To improve the efficiency of automatic test data generation, the multi-path test data generation method based on improved genetic algorithm is proposed. Firstly, the directed mutation operator was introduced into the genetic algorithm. The mutant was created according to the current best individual. It made individuals mutate in a good direction. The population diversity is maintained and the local search ability is improved. Secondly, the similarity degree of execution path and target path, and predicate branch distance were considered. An individual fitness evaluation function was designed. It distinguishes good individuals and bad individuals effectively. Experimental results on benchmarks show its superiority to the traditional methods.