In the central nervous system, more than a hundred white matter diseases have been discovered until now and they represent the second most frequent cause of death worldwide in 2012 according to official figures from World Health Organization. For the patient diagnosis, magnetic resonance imaging enables the development of emerging and non-invasive tools that better depict biological alterations in white matter. In pharmaceutical research, the implementation of these techniques would enable to assess more accurately novel therapeutic solutions. In this context, diffusion-weighted imaging offers a great opportunity to characterize non invasively white matter microstructure by diffusion compartment models. The diffusion of water molecules in biological tissue can be modelled by taking into account intra- and extra- cellular spaces. Based on this cellular representation, white matter microstructure comprised of axons, glial cells and extra-cellular space can be represented by multiple compartments. Different features such as axonal diameter or fiber dispersion have been highlighted by these diffusion models but few of them have been applied to characterize neurological diseases. In this thesis, diffusion compartment models have been investigated with animal Wallerian degeneration models on the basis of two main biological criteria: axonal loss and the presence of infiltrating cells due to immune response. Axonal loss is one of the first major alterations observed in nervous tissue for various neuropathologies. The relative quantification of intact axons is essential to describe the severity of the injury. On the basis of this quantitative criterion, several diffusion compartment models have been compared and bi-compartment models were better than three-compartment models. The chronic immune response with the presence of infiltrating cells is a second important feature observed in white matter diseases. Our Monte-Carlo simulations have shown that the presence of vesicles can be highlighted by the drop of isotropic diffusivity. On the basis of these two biological alterations, a diffusion model (DIAMOND) was selected to perform a longitudinal study on a Wallerian degeneration model. This model has shown significant correlations between histology and diffusion parameters for the characterization of axonal loss and the microglial phagocytosis.