To solve problems that exist in optimal design such as falling into local optimal solution easily and low efficiency in multi-objective optimization, a new approach based on design of experiments (DOE) and gradient optimization (GO) was proposed. The new optimization method is called DPG (DOE Plus GO) which used DOE for preliminary analysis of the function model, and took the optimal values obtained in DOE stage as the initial values of design variables in GO stage so as to reduce the effect on the result of optimization made by the designers� decision. This paper gave two typical examples of optimization to confirm DPG global, efficient, and accurate with Isight code. Firstly, the bimodal problem was used to test DPG�s global optimization ability, then the multi-objective optimization of the machine tool spindle, which required minimum quality, maximum stiffness, and strength was conducted. The results show the DPG optimization method could not only avoid falling into local solution, but also have an obvious superiority in treating the multi-objective optimization problems.