The martensitic phase transformation in steel is one of the most famous solid state transformation because of its contribution in the development of the modern industry. This transformation from a face-centered cubic (FCC) high temperature phase, called the austenite, to a metastable body-centered cubic (BCC) low temperature phase, the martensite, is crucial in the steels metallurgy, as the strength of the material relies on it. Despite numerous studies, this transformation exhibits particular features that are still not fully understood. Among these, the variant selection phenomenon, observed when stress and/or strain is applied to the material during or prior to transformation, is of particular interest due to its implications in the industrial processing of steels. Various criteria have been proposed in the literature to model variant selection, most of them being based on the work of the deformation associated with the transformation in the applied stress field. The deformation used in the computation of the work is generally determined by using the Phenomenological Theory of the Martensite Crystallography (PTMC). However, this theory considers different deformations to model the transformation, and there is currently a lack of consent on which of these different deformations is the most appropriate to account for the selection phenomenon. In this thesis, we propose to use a novel mechanistic transformation model, called the continuous distortion model to study the selection phenomenon. This model describe the transformation of the FCC lattice to a BCC lattice by using an atomistic description of the lattice change. The first part of this thesis is a study of the capabilities of the new model in accounting for crystallographic features of the transformation. Our study indicates that the crystallographic nature of the interfaces between the martensite and the austenite can be explained by combinations of different variants of the continuous model. It appears also that under the same assumptions, the mechanistic and the phenomenological model are equivalent. In the second part of this thesis, the variant selection is studied experimentally. Different thermomechanical treatments are applied to various types of steels, and the produced microstructures are characterized by Electron BackScattered Diffraction (EBSD). The experimental results suggest that the deformation that needs to be consider in the computation of the work depends of the type of strain accommodation mechanism involved in the transformation. When the accommodation of the martensite product is achieved by dislocations, the deformation associated with individual variant should be considered in the computation of the work. On the contrary, when the accommodation is based on variants coupling, it is the combination of the deformations of the coupled variants that matters. This thesis lays the foundations for a more unified vision of variant selection in martensitic steels, by considering a unique deformation model, and underlines the importance of the accommodation mechanisms in the selection phenomenon.