Using contextual factors beyond individual factors, contextual analysis allows a more accurate identification of at-risk populations, which could be useful when planning health programs. Multilevel models, widely used in British and North-American social epidemiology research but less frequently in France, are particularly suitable to analyse contextual data, because they take into account their hierarchical structure. This paper addresses methodological issues in the utilization of multilevel models, and reports some results which illustrate their potentials compared to those of more conventional statistical methods. As well as other methods, multilevel models are able to take into account the hierarchical structure of the data when estimating parameters. Furthermore, and more specifically, these models can also be viewed as useful tools to investigate contextual effects. Their particular interest is to disentangle individual-level variability and between-group variability. Comparing the group-level variance before and after introduction of individual-level characteristics allows to assess the extent to which between-group variability is linked to compositional effects. Multilevel models can also help examine whether the between-group variations affect all the members of the groups, or only specific sub-groups. Finally, they can estimate how much of this complex between-group variability is explained by the contextual factors included in the model. The overall conclusion is that multilevel statistical methods should be used in social epidemiology studies dealing with individual and contextual data, to produce results that are both richer and more consistent.