Abstract Network percolation is one of the core topics in network science, especially in understanding and optimizing the robustness of real-world networks. As a powerful tool, the message-passing approach shows unique advantages in characterizing network percolation compared with the mean-field approach. This approach simulates the behavioural response when the network is damaged by transmitting and updating messages between network nodes, thereby accurately assessing the robustness of the network. This paper reviews the progress of message-passing approaches in network percolation on simple networks, multilayer networks and higher-order networks in recent years and discusses the application of this approach in other research fields. Finally, we discuss future research directions around this approach.