Although scholars have established that voters have unstable preferences (e.g., Converse, 1964; Zaller, 1992) and that they are not accurate when recalling past preferences (e.g., Markus, 1986; Niemi, Katz, & Newman, 1980; Smith, 1984), existing research has not systematically explored whether voters can accurately predict the changing nature of their own opinions. The question of whether whether people recognize the instability of their political preferences was explored in a random sample of Pennsylvania registered voters who were surveyed in August and October 1996, during the presidential election campaign. The first survey elicited respondents' positions on two political issues (welfare reform and the environment) and on the two major candidates, and also asked them to estimate the likelihood that each of these positions would change during the next 2 months. The second survey elicited positions at that time and also asked voters to recall their prior positions. Measured both by expectations and recall, respondents tended to underestimate the degree to which their own positions would change or had changed over time. This research has implications for the use of public opinion polling and more broadly for the practice of democratic politics.