This application provides an isolation and recovery method and related network device for a case when one or more physical layer apparatuses (PHYs) in a flexible Ethernet group (FlexE group) are faulty. In the method, if a network device determines that a first overhead block corresponding to each current available PHY is stored in a corresponding memory, the network device determines that a FlexE group meets a PHY alignment condition, and starts to simultaneously read cached data from all memories. Therefore, there is no need to insert local fault LF code blocks to all clients, and there is no need to recreate a group. This effectively reduces the impact of a faulty PHY on client services carried by a normal PHY.