Reliability Optimization for FMS (flexible manufacturing system) is very important. Traditional optimal algorithms applying to FMS reliability have many shortcomings. In this article, the authors take into account the characteristics of FMS and bring forward modular Neural Networks (NN) model based on Genetic Algorithm (GA) to optimize FMS reliability. Modular NN based on GA has good convergence and short learning time. The example shows that the new algorithm is feasible and effective.