The preforming stage of the LCM composite manufacturing processes lead to fibrous reinforcement deformations which may be very large especially for double curvature shapes. Those deformations have significant influence on the second stage of the process, i.e. the injection of the resin. A way to predict accurately the spatial distribution of the permeability tensor consists in simulating for various configurations, the deformed shape of the reinforcement at the scale of the yarns. Mesoscopic scale analyses of textile reinforcements generally consider the yarns as a continuous material despite their fibrous nature. In order to have an accurate simulation tool, it is necessary to build up a constitutive law which accounts for the physical specificities linked to the microstructure of the yarns. Several models exist with reasonable accuracy. The present paper proposes a new approach in the hyperelasticity framework. The proposed model is based on the definition of mathematical invariants linked to the four main deformation modes of the yarn material: tension, compaction, longitudinal shear and transverse shear. The strain energy potential build up with those invariants is identified using classical fabric material tests: uni- and bi-axial tension and compression. The model has been validated on laboratory tests such as bias extension tests and gives promising results.