This paper presents a new modelling method for complex control systems. First, a new system model called a heterogeneous model is proposed. An identification method is given to identify the heterogeneous model. The identification method includes three algorithms; the recursive partitioning algorithm which optimally divides the entire domain of the system variables into a series of subregions, the dynamical structure selecting algorithm which selects the structure of the subsystem on each subregion, and the local model identification algorithm which gives the local analysis model of each subsystem. The entire identification is a recursive scheme which includes two procedures, a forward stepwise procedure to select the terms of the heterogeneous model that contribute sufficiently to the accuracy of the fit, and a backward stepwise procedure to remove the terms of the heterogeneous model that no longer contribute sufficiently to the accuracy of the fit. So the recursive procedure is guaranteed to produce the simplest model of the complex system.