Generally an error recovery module performs three activities: error detection, identification, and diagnosis. This paper presents an approach for dealing with several critical issues that arise in performing these three activities: error classification, error knowledge representation, and generation of recovery procedures. Execution errors occurring in a manufacturing system are emphasized in this paper. A formal model called MPSG (Message-based Part State Graph), which represents the execution activities of the shop floor controller, is applied to develop this approach. The objective is to simplify and generalize the error types and the parameters required by the error knowledge representation. The proposed method shows that this simplification and generalization provides a more convenient interaction with the subsequent diagnostic activity of generating recovery procedures. Moreover, it contributes to the integration of the error recovery module during the process of controller construction.