[en] The goal of this project is to develop and the validate a two-level calculational scheme suitable for the fast and accurate determination of nuclear cross sections with the lattice code DRAGON. Generally, classical one-level calculation schemes are used, employing the method of collision probability, which was the preferred option due to its relative simplicity, or the method of characteristics, capable of handling larger problems (more unknowns) leading to more accurate results. The problem is that a larger number of unknowns implies a longer calculation time. One solution to achieve both speed and accuracy is to combine these different methods in a two-level scheme. The purpose of this project is to validate an advanced two-level scheme with the use of the DRAGON lattice code, comparing it to classical one-level schemes, but also comparing the results with various other codes. The two-level scheme will be validated through two comparative studies on Pressurized Water Reactors (PWR) assemblies, this type of reactor being the most prevalent in the world. The first study will be carried out on an exercise dealing with typical 18x18 German assembly (KONVOI type). This study will aim to validate the scheme through various internal comparisons with the DRAGON code, and then comparing the results obtained with those coming from other codes. The second study will focus on a U0₂/MOX OECD full core benchmark. It will be an opportunity to set up a two-level scheme with DRAGON, to produce condensed cross sections libraries that will be used in the coupled code system QUABOX-CUBBOX/ ATHLET for a transient study on a complete a core, in comparison with libraries generated by the HELIOS code. (author)