To solve the problem of serious roller wear and improve the smoothness of the sliding door motion process, the rigid-flexible coupling multi-body model of the vehicle sliding door was built in ADAMS. Force boundary conditions of the model were determined to meet the speed requirement of monitoring point and time requirement of opening-closing process according to the bench test specification. The results of dynamic simulation agreed well with that of test so the practicability and credibility of the model was verified. In the optimization of the ride comfort of the sliding door, the shape parameters of the middle guide and the position of arm-shafts were selected as design variables while the impact load of rollers, the curvature of the trajectory and angular acceleration of the sliding door centroid were taken as optimization objectives. Firstly, eight simulation models with different parameters of middle guide were analyzed respectively and the model with the optimal ride performance was selected for multi-objective optimization. Multiple sets of sample models were obtained by using orthogonal experimental design and approximate surrogate models were established with the method of RSM and Kriging. However, the optimum solution couldn’t be obtained by using approximate surrogate models because the optimization objectives are highly nonlinear with respect to the design variables. To solve the problem, each model’s response values were analyzed and compared to find the lower response region which means the better ride performance. Generated the new sample models in these regions and obtained the convergence region of the response values. Finally, the response values of the optimization objectives were decreased by more than 20%. The proposed study improves not only the ride comfort of the sliding door, but also has great significance for the preliminary design and development of the sliding door.