The finite element simulation of dry fabrics forming has proved to be an interesting way in order to decrease the cost of the tools, predict the feasibility of a part with given tools, fabric and resin [1-3]. But these simulations requires accurate models of the mechanical behavior and permeability of the dry fabrics. Experimental methods are efficient to identify these properties but they need to be complemented with numerical approaches because they are time consuming and often very difficult to perform as far as composites are concerned and because they can not deal with the design of new fabrics is concerned. Numerical approaches at the elementary cell level permit to obtain accurate results on mechanical behavior and permeability for example[4]. But, these meso-scale studies requires an accurate 3D mesh of the fabric elementary cell [5]. Thus, we have proposed a consistent 3D geometrical model of woven fabrics [6]. The results obtained with this approach were interesting for 2D fabrics. Experimental observations have been performed in order to determine real yarn geometry in different cases of yarn structure and weaving. The analysis of these results helps us define the accurate generic 3D model of a yarn shape when it is weaved. Using this model of a yarn, a consistent 3D geometrical model of different fabrics is presented. The particularity of this model is that it ensures a realistic contact surface between yarns without interpenetration for all types of weaving. What is presented here, is an extension of this approach to 2.5D fabrics. The same strategy could not be used because of the great diversity pf patterns for 2.5 D Fabrics, that's why the approach is very different of the one that was used for 2D fabrics.