Abstract: Because the mode-tracking ability of a conventional adaptive grid interacting multiple model (AGIMM) is not strong, when the system mode has a long span jump, the model-set cannot adapt to the mode change quickly enough, resulting in mode tracking failure. In this paper, an improved AGIMM that mimics the antigen identifying mechanism of the immune system is proposed. In this algorithm, a monitor set is introduced to detect the distribution of the system mode randomly, which guides and corrects the adaptation process of the filtering model-set. Simulation results show that this algorithm has better mode tracking ability.