Abstract: Compared with the principal component analysis (PCA) method, kernel principal component analysis (KPCA) method has more advantages in the monitoring of nonlinear processes. However, it is difficult to find an inverse mapping function from the feature space to the original space for KPCA, resulting in great difficulties for the KPCA-based fault diagnosis. To solve this problem, the fault identification index was improved on the basis of KPCA fault data reconstruction method. The improved method could identify both univariate faults and multivariate faults. In addition, the proposed method could also reduce calculation and avoid the defect that the traditional fault detection methods could only identify univariate faults. The simulation results indicated the feasibility and effectiveness of the proposed method by testing it in the Tennessee-Eastman process.