Although there are some skillful techniques to analyze the parameter design problems, the methods for tackling the dynamic multi-response are rare. This work proposes an approach based on backpropagation neural networks and desirability functions to optimizing parameter design of the dynamic multi-response. A novel performance measurement of dynamic multi-response is developed to apply the desirability function that integrates several different types of dynamic responses into a single index. The proposed approach employs a BPN to construct the response model of the dynamic multi-response system by training the experimental data. The response model is then used to predict all possible multi-responses of the system by presenting full parameter combinations. Through evaluating the performance measurement of the predicted dynamic multi-response, the best parameter setting can be obtained by maximizing the single index. An illustrative example is analyzed to demonstrate the effectiveness of the proposed approach.