Aimed at the disadvantages of fuzzy C-means in fault diagnosis of stea m turbine set, a weighted fuzzy clustering method based on particle swarm optimi zation is put forward. Firstly, the method adopts similarity based weighting me th od to assign feature weights and sample weights in order to handle the variety of samples with complicated distribution. Then, the particle swarm optimization with compression factor is used to optimize feature weights and clustering target function of weighted fuzzy clustering. Finally, the best clustering num and clustering result are adaptively obtained by clustering validity function. Applicat ion results show that the method reduces the misclassification rate in fault dia gnosis of steam turbine set with the features of fast convergence and global con vergence. It can diagnose single fault and compound faults with high reliability and practicality.