To address the problem of low detection accuracy when the fault node rate was higher than 50% in Wireless Sensor Network (WSN), a wireless sensor fault diagnosis algorithm based on the precondition of neighbor nodes and neighbor node data was proposed. Firstly, the historical data of nodes were used to pre-calculate the states of sensor nodes initially. Then the final state of each node was judged by taking advantage of similarity of nodes and pre-states of neighbor nodes. Finally, the fault node information was sent to the base station by mobile sensors through the optimal path, which effectively reduced the number of communications. A WSN was simulated in an area of 100 m*100 m. The experimental results show that compared with the traditional Distributed Fault Detection (DFD) algorithm, the diagnosis accuracy of the proposed algorithm is improved by 9.84 percentage points. Moreover, the proposed algorithm even achieves more than 95% fault diagnosis accuracy when the node failure rate is as high as 50% in the network. In practical application, the proposed algorithm improves the fault diagnosis accuracy, reduces the energy consumption effectively, and prolongs the network lifetime as well.