Studies of retrospective voting in Great Britain largely focus on economic issues, and pay little attention to whether electors associate economic changes (both positive and negative) with government policy. This paper analyses voters' retrospective evaluations of nine separate policy areas (over three main policy domains) in 1997, showing that the more negative evaluations of government policy over the previous five years the less the likelihood of an elector voting for the incumbent party -especially if in addition the main opposition party is viewed favourably. This provides strong support for the reward-punish model. KEY WORDS .1997 general election. Britain. economy. public services. retrospective votingWith the increasing dealignment of the British electorate and the decline in importance of traditional social cleavages in accounting for variations in voting behaviour there, analysts have turned to alternative models. (On dealignment and the 'death' of the class cleavage, see Sanders (1997) andEvans (1999a).) Initially, analysts were attracted to models with a focus on issue voting, but this has been eclipsed by a concentration on rewardpunishment models according to which voters reward governments for positive policy outcomes and punish them for negative ones. ('Reward a governing party with your vote when satisfied, punish it by voting for the opposition when dissatisfied') (Norpoth, 1992: 57). These models -alternatively known as responsive-voter models -originate in classic texts by Schumpeter and Key. Schumpeter's (1942) seminal contribution equated