Through a series of 5 pre-election web-based studies, we randomly assigned respondents to either a bipolar or unipolar presidential candidate vote intention scale (with the item stem being 'How likely are you to vote for George W. Bush?' and 'How likely are you to vote for John Kerry?'). We compared the likelihood measures with a categorical candidate preference measure ('Who would you vote for?') to determine concurrent validity. We then recontacted these respondents following the election to inquire who they voted for to determine the predictive validity of these measures. We compare our results with previous research that indicated that unipolar behavior intention measures (i.e. "Not at all likely" to "Extremely likely") have somewhat higher validity than bipolar behavior intention measures.