The model-based fault diagnosis approach is characterized by a powerful process supervision capability with a priori knowledge about the system under consideration. Nevertheless, system complexity, high dimensionality, process nonlinearity and/or lack of good data often hamper its application in chemical engineering systems. A nonsteady state model based fault detection and diagnosis method for the distillation process was developed, using dynamic simulation to monitor the distillation process and identify abnormal sources when large deviations among measuring variables occur. It continuously updates the inner distillation parameters via on-line correction and predicts the trend of measuring variables and determines the existence of malfunctions simultaneously. The distillation model is dependent on transfer equilibrium, mass and heat balance, and is simulated by Euler and two-tier approach. This method was demonstrated with simulated data of a stripping tower collected from the Tennessee Eastman chemical plant simulator.