Aimed at the problem of the multiple fault diagnosis of redundant system, a multiple fault diagnosis strategy for redundant system based on probability is proposed to reduce the computation and the expectant costs under condition that a probability value is set as a probability threshold and the multi-fault state whose probability is lower than the value of this probability threshold is neglected. First, this strategy is used to optimize the relation matrix model, and give up the multi-fault states whose probability is lower than the probability threshold. And then, Rollout algorithm is used to optimize the information entropy based diagnosis strategy. The example demonstrates that though the expectant costs are lowered by the strategy into a certain accuracy, a great proportion of the expectant test costs is reduced ,the size of the fault-test relation matrix model is minimized, and the calculation time is reduced.