[en] The numerical solution of the transport equation is greatly facilitated by using a multigroup discretization in energy. Multigroup discretization comes along with resonance self-shielding, which is one of the main error sources in reactor physics. The aim of this project is to validate a resonance self-shielding method developed for the DRAGON code in the Institut de Génie Nucléaire at the École Polytechnique de Montréal : the Subgroup Projection Method (SPM). This project is related to the development of fourth generation fast neutron reactors, for which the SPM method has never been used. Two different aspects of the SPM method have been studied : 1. the SPM method makes a series of assumptions called the “Livolant-Jeanpierre approximation”, which allows to decouple the self-shielding calculation and the main flux calculation. On the contrary, in the European code ECCO, the self-shielding calculation and the flux calculation are performed together. Thus, the results given by DRAGON and ECCO have been compared. Preliminary studies have been carried out in order to suppress difference sources not related to the Livolant-Jeanpierre approximation. The study proves the validity of the Livolant-Jeanpierre approximation for any kind of neutron spectrum and thus the possibility to use the SPM method for fast neutron reactors. 2. in the SPM method, every cross section is condensed with the scalar flux. This project aimed to calculate the errors generated by this assumption, particularly when leakage is involved. The ECCO code has been used again, because contrary to DRAGON it performs a specific condensation for the total cross section and the anisotropic diffusion cross section. According to the results of this study, the total cross section should not be merely condensed with the scalar flux whenever a leakage model is involved. The impact of the condensation of the anisotropic diffusion cross section appears to be insignificant. (author)