A structure optimization method of machine tool weak part based on mapping model between structure and whole machine performance was proposed. In this method, firstly the structure weak component was determined by the dynamic and static characteristics analysis of machine tools. Secondly, the structure-performance mapping modeling method based on elliptical basis function (EBF) neural network, whose extended constant was selected adaptively, was proposed. In this section, the elliptical basis function neural networks was modified and improved, and the EBF modeling method based on self-adaptive extended constant was proposed. The self-organizing selection of expansion coefficients was used to determine the reasonable participation and overlap of different elliptic basis functions, and it can avoid all elliptical basis functions from too flatting or too slant effectively, which may affect the accuracy of EBF modeling. Then, the structure-performance mapping model based on improved elliptic basis function neural network was structured. Also the validity and correctness of the mapped model was verified based on the sample data: the correlation coefficients between actual values and calculation results from mapped model were all above 0.995. Thirdly, on the above basis, according to the physical mapping relation between structure and static/dynamic performance of the whole machine tool, considering the effect of boundary constraint of the whole assembly, by taking dynamic and static performances as evaluation, and choosing the structure of weak component as the optimization object, based on multi-objective optimization algorithm, the optimization of weak structure part and the whole dynamic performance of machine tool were realized finally. After optimization, the center point deformation of tool was reduced by 12.8%, the mass of structure part was reduced by 9.7%, while the first order natural frequency of the whole machine tool was increased by 6.9%.