In most democracies, at least two out of any three individuals vote for the same party in sequential elections. This paper presents a model in which vote-persistence is partly due to the dependence of the utility on the previous voting decision. This dependence is termed habit formation. The model and its implications are supported by individual-level panel data on the presidential elections in the USA in 1972 and 1976. For example, it is found that the voting probability is a function of the lagged choice variable, even when the endogeneity of the lagged variable is accounted for, and that the tendency to vote for different parties in sequential elections decreased with the age of the voter. Furthermore, using structural estimation the effect of habit is estimated, while allowing unobserved differences among respondents. The structural habit parameter implies that the effect of previous votes on the current decision is quite strong. The habit model fits the data better than the traditional party identification model.