Models of government formation processes in multi-party democracies are unusually highly sensitive to the rules that govern the selection of formateurs. The formal literature has exclusively focused on two selection rules: selection proportional to seat share (Baron and Ferejohn 1989), and selection in order of seat share (Austen-Smith and Banks 1988). In this paper we use a new data set on government formation attempts in 12 multi-party democracies to empirically assess which selection rule most closely approximates the data. We find that while there is no empirical support for selection in order of seat share, proportional selection may serve as a good first approximation. However, our results also indicate that formateur selection exhibits a significant incumbency effect: ceteris paribus, the former prime minister's party is more likely to be selected to attempt the formation of the next government.