Regarding application with smooth variation of detection, spatial correlation of sensors’ data within a small field was applied to sensor nodes’ fault diagnosis. The data were sorted into several continuous sequences by sink node. Sequence with minimum variance was regarded as normal data to determine normal nodes. For undetermined nodes, it can be determined via calculation on deviation to normal nodes’ data of vicinity area. If deviation does not exceed the threshold, the node is normal; otherwise, it is regarded as a fault node. The research on WSN in a greenhouse shows that fault node can be effectively detected in time by this method.