Abstract: A structural optimization design method based on an RBF neural network model and combined optimization strategy was proposed in order to obtain the explicit functional relationship equation between design variables and performance objectives in the structural optimization of complex machine tools. Taking a type of precision lathe spindle box as the research object, the optimal Latin hypercube experimental design and sensitivity analysis of the spindle box design dimensions was carried out using the joint simulation technology of finite element software ANSYS Workbench and multidisciplinary optimization software Insight and the RBF neural network model was constructed in addition to the spindle box finite element model according to the experimental sample points. A combined optimization strategy combining multi-island genetic algorithm (MIGA) and sequential quadratic programming method (NLPQL) is used to optimize the design of the RBF neural network model. The optimization results show that the mass is reduced by 12.89% while the static and dynamic performance of the spindle box is guaranteed, which achieves the expected results in a more efficient way than the standard optimization.