Asphalt rubber is a modified binder widely used in pavement construction. Composition design methods of asphalt rubber blends are exclusively empiric and rely on the selection of the optimal combination of parameters after a long experimental program. The understanding of the phenomena affecting the asphalt rubber behaviour starts with the proposal of a rational theory and its validation with experimental data. To do so, a set of experiments were carried out including three main factors considered as significant for the rheological behaviour of the final binder. These factors are the bitumen chemical composition, the rubber content and its maximum mesh size. The binder complex shear moduli were measured within the viscoelastic linear domain. The measurements allowed a full characterisation of the rheological behaviour of asphalt rubber for the service temperatures and rates of loading. Base bitumen composition leads to final binders with rheological behaviours that are radically different despite the fact that the base bitumen belong to the same commercial grade. The rubber content increase showed significant changes on the properties of the binder, especially an increase of its elasticity and a decrease of its temperature sensitivity. The rubber powder maximum mesh size has no significant effect on the rheological properties of the blends within the service conditions. An analysis of physical and chemical changes was carried out on the phases formed after the curing process. A method for the isolation of the phases without affecting their physical and chemical properties was proposed. A centrifuging process was applied to asphalt rubber thanks to the design of an appropriate equipment adapted to the binder features. An experimental protocol was established to obtain a good separation of the phases with a limited aging. The SARA bitumen group analyses were carried out before and after the curing process. They allowed the determination the bitumen fractions kinetics during the blending. Light fractions of bitumen are the main chemical groups that migrate toward the cross-linked rubber network causing its swelling. The rheological behaviour of the separated bitumen phase has proven that the residual bitumen exhibits a lowered thermal sensitivity which is confirmed by the increase of the colloidal instability index. The rubbery phase has showed an extent of swelling varying between 200 and 400% which modifies considerably the rheological behaviour of the rubber by transferring to it a part of the thermal sensitivity of the bitumen. The influence of the factors was evaluated and their relative importance determined through a statistical model. Micromechanical modelling of the asphalt rubber composite has allowed the validation of observed phenomena that are mainly physical ones. To do this, two main homogenisation procedures were implemented and carried out on the heterogeneous medium: The classical self-consistent approach widely referenced as the equivalent medium approach (EMA) and the generalised self-consistent approach frequently referenced as the equivalent field approach (EFA). A satisfactory modelling of the composite is achieved by taking into account two changes occurring in the phases: their volumetric fractions influenced by the swelling, and their rheological properties due to the diffusion of bitumen into the rubber. An acceptable estimation of the rheological behaviour of the rubbery phase is obtained from the base materials properties with a three-phase self-consistent modelling according to EFA approach. The consideration of an adsorbed fraction of the bitumen on the surface of the rubber particle has allowed a qualitative and a quantitative description of the properties of the rubbery phase. The difficulty of predicting the rheological behaviour of a bitumen from its chemical composition is discussed. Future investigation ways are proposed.