Abstract: Fault-tolerance TCP is a critical technology for implementing services’ availability running on the server that is the necessary function of servers nowadays. A fault-tolerant TCP scheme （MI TCP） based on multi-images is discussed in this paper. The scheme makes each TCP connection have several synchronous connection images and doesn’t backup the status of TCP connections, which needn’t reincarnate a connection to start from the beginning. It issues a module mechanism in the kernel of Linux and doesn’t affect software running on both a client and a server. It also guarantees that each TCP connection can be taken over seamlessly when it fails and this process is transparent to the user. At the same time, the experimental results demonstrate that it can also try to reduce the effect of system performance on the premise of ensuring the fault tolerance of each connection. That is to say, MI TCP is proved to be efficient.