The model of redundant multiagent system (MAS) as distributed hardware-software complex, which is based on replication of tasks instead of replication of agents, is developed. The developed fault recovery methodology presented in the article includes communication protocol for components of MAS, which enables task execution independently of its deployment or belonging to a particular component of MAS, technique for keeping inactive task replicas in actual state and techniques for fault detection and fault recovery. Developed models and methodology are validated by stated and proved theorem on fault tolerance property of MAS.