A model-based three-way automotive catalyst monitoring and fault detection strategy is presented in this work. The performance of the catalyst is inferred from the error between the post-catalyst exhaust gas oxygen sensor air fuel ratio measurement and the model predicted value. A simplified catalyst oxygen storage and reversible deactivation model is employed to predict the post-catalyst air fuel ratio. The model-based strategy is based on the use of a test statistic that is computed from a window of post-catalyst air fuel ratio prediction error data updated in real-time. A fault is assumed to be present when the value of this statistic exceeds a threshold determined from some specified confidence level. We illustrate several test statistics using engine operating data and conclude with an evaluation of this strategy. (c) 2007 Elsevier Ltd. All rights reserved.