Building on the pioneer work of Lycan (1988, 1991a and b, 1994) and Yagisawa (1988), this paper aims to set out in a comprehensive way the case for the view that Lewis' analyses of modality in terms of quantification over worlds and counterparts fail to be genuinely reductive. This involves bringing together elements of the case which have hitherto remained unconnected, motivating some of those elements in new and more decisive ways, and bolstering them by considering a range of objections that have been levelled against them (including recent ones by Stalnaker and Divers). The paper proceeds as follows. Section 1 introduces Lewis' analyses of de dicto and de re modality. Section 2 argues that, as deployed by Lewis, the notion of ‘world’ is modal in nature. This is because the analyses he proposes are extensionally adequate only if he takes ‘world’ to mean possible-as-opposed-to-impossible world. An important response to this charge is simply to deny that there are impossible worlds. Section 3 considers and rejects Lewis' argument for that claim, while section 4 argues that a prima facie case can be made for the claim that Lewis is in fact committed to the existence of impossible worlds.