In this paper, the local linear models of a magneto-rheological (MR) damper are obtained based on the Takagi-Sugeno (T-S) fuzzy modelling approach. In these local linear models, the output force of the MR damper is expressed as the linear summation of the state variables (relative displacement and relative velocity) and input voltage. To obtain these local linear models with high accuracy, the genetic algorithm (GA) with a new encoding method is applied to search for the optimal model parameters. The proposed hybrid intelligence technique can evolve the fuzzy rule structure (number of rules and selection of rules) and the input structure (number of premise inputs and selection of premise inputs) simultaneously so that the obtained linear models have the simplest structures without decreasing the modelling accuracy. To validate the proposed approach, the modelling errors between the MR damper output and the corresponding linear model output are compared for the given number of rules case and for the automatically selected rules case with using three different selection approaches for the premise input variables. It is confirmed by the validation results that the proposed hybrid intelligence technique can find the optimal linear model for the MR damper.