The development of inexpensive metal-nitrogen-carbon (M-N-C) catalysts for electrochemical CO2 reduction reaction (CO2RR) on an industrial scale has come to a standstill. Although the number of related studies and reviews has grown fast, the complexity of the M-N-C composite has limited researchers to focus on only a few variables and carry out sluggish trial-and-error optimizations in their studies. As a result, the conclusions are drawn only by artificial analysis based on a few orthogonal experimental results. To obtain more general design strategies, we have innovatively introduced machine learning (ML) into this field to address this bottleneck. A standard workflow that comprehensively utilizes different ML algorithms and black-box interpretation methods is proposed for this purpose. Besides predicting CO2RR performance metrics for M-N-C catalysts, such as maximum faradaic efficiency with great accuracy, the ML models have also indicated simple and clear design strategies that would guide future exploration from a data science perspective. Besides, we have also demonstrated the potential of the models in guiding the development of new material systems. We thereby believe that the new research paradigm proposed may accelerate the development of this field soon.