Second order nonlinear macroscopic motorway flow model is in the focus of the paper. First, the pure nonlinear traffic flow model is transformed to an approximate Linear but Parameter Varying (LPV) state space form. Second, this paper shows the efficiency of an LPV subspace based identification algorithm in closed-loop, the Predictor Based Subspace IDentification technique for LPV systems. Comparison of the results with the traditional traffic oriented identification (nonlinear parametric solution) is given. Models are validated using real loop detector measurements. Index Terms— Linear Parameter Varying (LPV) systems, LPV Identification, freeway traffic modeling I. INTRODUCTION In freeway traffic flow modeling, one of the most impor- tant question is how to properly describe or approximate the real traffic flow. The description can be derived from first principles models respectively from dynamics identified using simulated or preferably measured data sets. First, traffic theory gives alternative macroscopic approaches and therefore provide different, but first principle based models. Second, various identification tools might be applied to result in identified models. A widespread approach is to apply the macroscopic modeling principles, where the individual vehicle dynamics are neglected. Since the appearance of the fundamental paper (9), second order macroscopic models are in the focus of transportation research (8), (4). In these type of models, the traffic flow is described by two macroscopic quantities: vehicle density ρ (x, t) and (space mean) speed v (x, t), with x being the coordinate along the freeway and t being the time index in continuous time. Comparative works emphasize the benefits of the application of the second- order model describing the real traffic phenomenons more precisely, than the lower order systems (7). The differential equation of speed evolution (also called momentum equation) contains several unknown model parameters. These parame- ters are used to tune the model in order to adapt it to the real traffic conditions. Without the proper selection of parameter values traffic models can not be used in traffic estimation,