This paper aims at presenting a modelization of the uncertainty on the linear model obtained from the so-called linearizing feedback of dynamic flat systems. After such a linearizing feedback is computed to ensure good nominal tracking performance, a compact set of models can be computed by taking account of both state space disturbances and parametric errors. A robust linear feedback is then designed in order to guarantee a specified level of performance. This representation of the compact set of models makes use of the Linear Fractionnal Transformation, so that the global robustness of tracking performance is assessed by µ-analysis.