A fault-tolerant messaging system comprises a primary broker, a first network, a back up broker, and a second network. The primary broker and the back up broker are coupled to the first network for communication with clients thus creating a messaging system. The primary broker and the back up broker are also coupled to the second network for replicating state from the primary broker and the back up broker, and also sending transaction events immediately to maintain synchronization. The brokers further comprise a replication module for communicating state between the primary broker and the back up broker, a recovery module for performing recovery on the back up broker upon failure of the primary broker, and a fault-tolerant connection module for establishing a fault-tolerant connection between the primary broker and the back up broker over the second network. The present invention also includes a number of novel methods for performing fault tolerance.