An experimental approach to study the distribution of power in a voting body is described. Laboratory experiments in collective decisionmaking provide a relatively new way of measuring voting power that 1) is based on empirical data, 2) is analytically tractable and 3) allows to control voters’ preferences in a theoretically unambiguous environment. Taking the experiment by Montero, Sefton and Zhang (2008) (MSZ) as starting point, we conrm their basic ndings, and explain some of their empirical paradoxes. In particular, we show that the asymmetry between the resulting shares of some otherwise identical players is a feature of the experimental design of MSZ, and develop an experimental way to correct for this. The main contribution of our research deals with the question of how voters’ preferences to coalesce inuence their behaviour and the resulting allocation of shares (as measured by the average payos of the players). To tackle the issue experimentally we extend the basic design to allow for asymmetric voters’ preferences depending on the coalitions they take part in. The results of the experiments show that even small modications of preferences lead to statistically signicant dierences in players’ shares. This result supports preference-based power indices as proposed in Aleskerov (2006), rather than the classical Banzhaf and Shapley-Shubik indices.