Multi-component natural products provide both beneficial and toxicity effects to the health of human. However, the optimal combinations of natural ingredients to enhance the health beneficial efficacy and reduce toxicity problem has been recognized as a significant research issue in a variety of areas in pharmaceutical industry. Therefore, the primary motivation of this paper is to propose a systematic method based on robust design methodology in order to determine the optimal combinations and the associated amounts of multiple components natural product. First of all, the simplex lattice mixture design was proposed to explore the information between natural ingredients (i.e., input factors) and the associated toxicity and synergy output responses. Then, the stepwise regression and mixture screening design were performed to identify the significant input factors. Secondly, the neural network-based estimation method was used to estimate the functional relationship between significant input factors and output responses. Thirdly, the optimal input factor settings can be obtained by using desirability function model. Finally, a case study of green tea catechins was performed for illustration purpose.