Abstract: For the purpose of performance assessment and monitoring of MPC(model predictive control),a data-based covariance benchmark is adopted in performance assessment and monitoring.Generalized eigenvalue analysis is used to extract corresponding better and worse eigenvectors based on the outputs of the monitored period and benchmark period.The confidence intervals of the eigenvalues on corresponding directions are obtained by using statistical inference method.The performance indices within the isolated better and worse performance subspaces are then derived to assess and monitor the performance of MPC.This measure is successfully applied to performance assessment and monitoring in the Wood-Berry tower process with three mismatch cases.The simulation results show the feasibility and effectiveness of the method.