Considering that multiple faults usually exist together in a complex system and false alarm or omission may occur during test process, a multiple fault location method for complex system with unreliable tests is proposed. The problem of finding fault hypothesis with maximal posterior probability is transformed into a combinatorial optimization problem by using Bayes theory, the detection dependency matrix and false alarm dependency matrix. The method provides a modified reflection subgradient optimization algorithm (MRSOA) to solve the Lagrange dual of the optimization problem and obtain the fault location result. The result of experiment on multiple fault location for a radar transmitter with unreliable tests demonstrates that the proposed method has high location efficiency and diagnosis accuracy.