Abstract: To overcome the disadvantages that irrelevant variables spoil classifiers and decrease the correct classification rates of faults,a new optimization algorithm based on discrete particle swarm optimization(PSO) and support vector machines(SVM) is presented to directly search for fault feature variables.As it can reduce the dimensionality of data space and preserve the fault features,the algorithm greatly improve the performance of fault diagnosis.The simulation results of fault diagnosis on continuous stired-tank reactor(CSTR) prove its validity.