This paper illustrates what the use of multilevel models, applied to aggregate data (electoral results, sociodemographic and tax data), can add to the knowledge of votes for the French far-right party Front national. It underlines the link between methodological choices and substantial or theoretical conclusions. In particular, the hypothesis defended in the paper is that this methodology leads to looking at Front national votes through their spatial and temporal diversity, rather than as a unified reality with a univocal meaning. Moreover, this methodology leads us to consider the vote as a scientific object that can, and should, be considered in a relational, and not only simply individualistic (rational choice or psychosocial theories) or holistic (very general theories, putting the stress on large socio-historical changes) view. Thus, the reflexion on relations between individuals and context rests on the hypothesis that these two notions should be understood in their complementarity rather than as two separate realities, as Elias demonstrated.