Analyzed here is a probability learning fault localization algorithm based on directed graph and set-covering. The digraph is constituted as following: get the deployment graph of managed business from the topography of network and software environment; generate the adjacency matrix (M a ); compute the transitive matrix (M a 2 ) and transitive closure (M t ) and obtain dependency matrix (R). When faults occur, the possible symptoms will be reflected in R with high probability in fault itself, less probability in Ma visit check out, much less in M a 2 and least in M t . MCA+ is a probability max covering algorithm taking lost and spurious symptom into account. DMCA+ is dynamic probability updating algorithm through learning run-time fault localization experience. When fail to localize the faults, probabilities of real faults will be updated with an increment. The simulation results show the validity and efficiency of DMCA+ under complex network. In order to promote detection rate, multi-recommendation strategy is also investigated in MCA+ and DMCA+..