Compact engine models often consist of a set of nonlinear differential equations which predict the time development of the mean value of the engine state variables (and perhaps some internal variables): such models are sometimes called mean value engine models. Currently a great deal of attention is focused on constructing such continuous time models and on finding their parameters. This paper shows, that it is possible to identify an engine model from a linearized version of a mean value model for a CFI four-cycle spark ignition (SI) engine. Such an approach is useful because it preserves a physical understanding of the engine throughout the identification stage. Afterwards the identification results are available for general dynamic engine studies. The identfication techniques discussed in this paper include classical methods (step response) as well as modern statistical methods (Kalman filtering and Maximum Likelihood estimation). These techniques have been applied to a four cylinder SI engine. The results include an identification of the most important parameters and time constants of the engine. These are of interest for the construction of engine simulation models, for control studies and condition monitoring applications.