In general, a system can be modelled by using different approaches like the theoretical, the semi-empirical or the empirical one. A solution for solving this problem is to construct approximation models, whose behaviour is close that of the real system. The paper discusses aspects of the numerical modelling of a turbocompressor, used in fuel cell applications, and presents a comparative study. The first two models, one a semi-empirical approach and the other based on moving least squares algorithm (MLS), are considered as basic models. These basic models are used to develop a hybrid model of the turbocompressor. Thus, the paper begins with a short survey of compression FC subsystems and continues with a presentation of some approaches for compressors modelling. Then a semi-empirical modelling approach method, based on theoretical equations, tuned on experimental dataset and a numerical (empirical approach) based on the MLS algorithm, tuned on the same experimental dataset, are described. The results are analyzed and compared and finally an adapted MLS method, for modelling the characteristics of the turbocompressor, is proposed and is discussed. The result is a predictive approximation model, composed of two sub-models, which describes better the global behaviour of the turbocompressor and it can be used for on-line or off-line identification of the compressor. (C) 2008 Elsevier Ltd. All rights reserved.