Through a series of web-based studies conducted in 2000 and 2004, we examined the efficacy of a variety of vote intention measures. Specifically, in 5 primary elections in 2000 (NH, CA, OH, GA, NY), in the 2000 general election, and in the 2004 general election we examined numerous variants of vote likelihood. In all studies we recontacted respondents within a few days of the election to determine whether or not they voted. Some vote intention measures appeared to be superior to others and we discuss the implications for determining appropriate likely voter models in pre-election polls.