This paper proposes a method that automatically generates software test cases based on a genetic algorithm and mutation analysis. Our method combines random generation and refinement. Each test case is generated randomly in the first step, and then a set of test cases is refined by the genetic algorithm. To measure the adequacy of the test case set, we use mutation scores, which are based on the mutation analysis of software testing. Our proposed method, which is applied to a C programing language program, automatically generated test case sets with 100% branch and boundary value coverages. The generation time of one test case set was approximately 130 ms. © 2012 IEEE.