Abstract: Network diagnosis problem aims to obtain compatible fault mode which c an explain symptoms by a set of actions. Some diagnosis models have been propose d, but their descriptions of the problem with dependent actions were not accurat e enough and the results are not very optimal. A DBN(diagnosis Bayesian network) model was presented that consisted of symptoms nodes, fault hypothesis nodes, d ia gnosis action nodes and observation nodes. It combined the general Bayesian netw ork and the requirements of fault diagnosis. Under the assumption of independent diagnosis process, a fault diagnosis algorithm based on DBN model was proposed. The algorithm took dependent actions into account. Observation nodes were introduced to achieve lower diagnosis cost. Experiments show that the fault diagnosis method based on DBN can reduce the diagnostic cost effectively and sol ve diagnosis problem under dependent actions condition preferably.