This paper presents a distributed data fault detection scheme for wireless sensor networks. Faulty sensor nodes are identified based on the similarity measure among neighboring nodes and dissemination of the decision made at each node. This scheme is designed for the application scenarios that both temporal and spatial correlations of sensor data among neighboring sensor nodes are observable. The performance of this scheme is evaluated in Castalia simulation environment, in which senor data faults and physical process are elaborately modeled. Simulation results show that the proposed scheme performs well for two kinds of typical sensor data faults. Threshold setting and selection of similarity function in similarity test are highlighted as key sub-issues for the proposed detection scheme.