A new multi-response optimization method,combining seemingly unrelated regression (SUR) models with factorial effect principles,is proposed to solve the quality design problem with multiple correlated responses. The proposed method not only identifies significant variables for each response model with factorial effect principles,but also measures the quality level of all responses whose process capabilities meet the corresponding requirements by using multivariate process capability index. In addition,the proposed method considers the influence of the model parameter uncertainty and the predicted response variability on the optimization results through Bayesian sampling technique. Firstly,a binary variable indicator is set for each variable in the SUR model to account for the factorial effect principles. After that,mixed binary variable indicators are constructed to improve the functional relationship between process responses and experimental factors. Thirdly,significant variables are identified by calculating the posterior probabilities of mixed binary variable indicators and different model forms. And the optimal model form is determined based on the above results. Fourthly,based on the previous steps,the optimal parameter settings are found by using multivariate process capability index. Finally,the results of two practical examples demonstrate that the proposed method can not only effectively identify significant variables for multiple related responses,but also provide the optimal parameter settings.