The approximate model of the muzzle brake performance was set up and evaluated to simplify the analysis process. LHS(Latin Hypercube Sampling) and numerical simulation of inviscid muzzle flow field were applied to obtain some samples of the muzzle brake performance. The performance was weighted with the impact force on the muzzle brake. Then RSM(Response Surface Method) was adopted to get the approximate model of the muzzle brake performance to establish a mapping of muzzle brake shape parameters and the impact force. In the end GA(Genetic Algorithm) was applied to perform the optimization of the muzzle brake shape parameters with the approximate model.