Aiming at the optimum design of thick composite structure which is formed by repeating sub-laminates,?a multi-step optimization method is proposed in this paper. This method is composed of two steps. The first-step optimization design is conducted using the approximate model which is based on radial basis function neural network. The design variables include the number of plies in the sub-laminate and ply volume proportion. The second-step optimization is divided into two levels. The design variables of the system level are the number of sub-laminates, and the objective is to minimize the structural mass. The sub-system level is the optimization of laminate stacking sequence using genetic algorithm, and the objective is to minimize the inter-laminar stress factor. Matlab is adopted to program the genetic algorithm, and then integrated with Isight to carry out the whole optimization process. The case study shows that the multi-step optimization design method proposed in this paper is effective. The optimization method provides the idea for the initial design of thick composite structure and is of significant practical value in engineering.