The fusion task is to merge document lists retrieved from a corpus in response to a query. Despite the few decades of research work on fusion, there are still no theoretical tools to formally analyze fusion methods. We use the Condorcet voting rule as a first step to this end. We then present an in-depth empirical analysis of TREC runs and existing fusion methods with respect to Condorcet. Additional exploration shows that a simple approach to turn existing fusion methods to Condorcet methods yields highly effective performance.