The basic principles of wave-front sensing based on phase diversity (PD) are expatiated. An improved PD based on genetic algorithm (GA) is proposed for the highly nonlinear character of PD′s objective function. The parallel and global search features of GA are suitable for the global optimization of PD by analyzing the characters of GA. A sparse-aperture system composed by three sub-apertures with different piston errors is established. The changes of the optimization objective function are analyzed and the new PD based on GA is used to find the global optimal value of the system. The results of many experimental simulations show that:GA can find the global optimum value of the sparse-aperture system,and the root mean square (RMS) between estimated piston error found by GA and the true piston error is 0.04λ. The new PD has a high accuracy.