We argue that voters' assessments of party leaders are comparative and prospective rather than individual and retrospective. Therefore, a prospective leadership-comparison evaluation should outperform a leader-approval, retrospective indicator as a determinant of government and party popularity. Using data from 1984–92, a popularity function that includes a variety of economic and political components, and several dependent variables, we test this hypothesis by comparing the performance of a ‘best prime minister’ question and the more usual ‘approval’ questions about party leaders. We find that the former gives consistently better results than the latter.