Abstract: The sensor node usually randomly locating in the harsh environment or the area no one can touch, with the nodes easily having fault. Therefore, the sensor node diagnosis method based on KMeans and rough set neural network was researched. Firstly, the improved KMeans algorithm was used to get the discrete data of the continuous attributes, then the rough set mutual information method was used to realize attribute reduction to improve diagnosis efficiency. Finally, the threelayer network fault diagnosis model was established, and the leapfrog algorithm was used to further optimize the weights to get the final fault diagnosis model. The simulation results showed that the proposed method can achieve fault diagnosis for sensor nodes, and compared with the other methods, it has the higher diagnosis precision and less diagnosis time.