The catalyst layer (CL) is the core component in determining the electrical-thermal-water performance and cost of proton exchange membrane fuel cell (PEMFC). Systemic analysis and rapid prediction tools are required to improve the design efficiency of CL. In this study, a 3D multi-phase model integrated with the multi-level agglomerate model for CL is developed to describe the heat and mass transfer processes inside PEMFC. Moreover, a research framework combining the response surface method (RSM) and artificial neural network (ANN) model is proposed to conduct a quantitative analysis, and further a rapid and accurate prediction. With the help of this research framework, the effects of CL composition on the electrical-thermal-water performance of PEMFC are investigated. The results show that the mass of platinum, the mass of carbon, and the volume fraction of dry ionomer has a significant impact on the electrical-thermal-water performance. At the selected points, the sensitivity of the decision variables is ranked: volume fraction of dry ionomer > mass of platinum > mass of carbon > agglomerate radius. In particular, the sensitivity of the volume fraction of dry ionomer is over 50% at these points. Besides, the comparison results show that the ANN model could implement a more rapid and accurate prediction than the RSM model based on the same sample set. This in-depth study is beneficial to provide feasible guidance for high-performance CL design.