A neural network alpha-beta-gamma filters optimized by an improved genetic algorithm (GA) was presented. In this new algorithm, a special fitness function on the basis of the tracker performance and adapted crossover and mutation probability were designed. So that premature convergence can be avoided, and the population diversity can be maintained. The improved GA ensures that the obtained parameters are optimal. And the proposed method provides a design approach for alpha-beta-gamma filter optimization to nonlinear path. Simulation results show that the improved algorithm possesses satisfied performance and strong robustness.