Second-order election models are based on several assumptions about individual-level motivations. These can be summarized by a transfer hypothesis: individuals presumably apply their evaluations of national-level phenomena to the EU level when voting in EU elections. In contrast, a suis generis hypothesis stipulates that voters evaluate the EU on its own performance terms. This paper tests these competing hypotheses. We find considerable support for both models. In the election context, where national institutions — political parties — dominate the representation process, the transfer hypothesis receives considerable support. However, we also find surprisingly strong support for the first-order hypothesis: electoral choice in EU election is influenced to a considerable extent by EU level factors. Furthermore, when voters evaluate the mechanisms of representation more broadly without a focus on elections per se, we find much more support for the first-order than the transfer hypothesis — voters clearly separate the two levels and evaluate each level on its own terms. These results have important implications, both for how we analyse voters' decisions in European elections, and how we view the sophistication of voters more broadly in the context of multi-layered institutions. [ABSTRACT FROM AUTHOR]