Résumé en anglais

Preferences can be aggregated using voting rules.We consider here the family of rules which performa sequence of pairwise majority comparisons between two candidates. The winner thus depends onthe chosen sequence of comparisons, which can berepresented by a binary tree. We address the difficulty of computing candidates that win for sometrees, and then introduce and study the notion offair winner, i.e. candidates who win in a balancedtree. We then consider the situation where we lackcomplete informations about preferences, and determine the computational complexity of computing winners in this case.