The study appears today in the journal Science.
Timothy Wilson, a psychology professor at the University of Virginia in Charlottesville and the author of the book Strangers to Ourselves: Discovering the Adaptive Unconscious, said there are two interpretations of the study results.
"It may be that participants really had made up their minds and just didn't know it yet," said Wilson, who was not involved with the study but wrote an accompanying article in this week's Science.
"Or they may have been leaning in one direction unconsciously and that biased how they interpreted the information they got about the issue in subsequent days."
Gawronski, the study co-author, says automatic mental associations play a particularly important role in a person's decision-making when it comes to ambiguous situations, such as political debates.
"In a debate between Obama and McCain, it may not be entirely clear who showed the better performance," he said.
"But undecided voters with more favorable associations with McCain may see him as the one who did the better job" and vice versa, Gawronski said.
"It's this biased perception of events that then provide the basis of their future decisions," he added.
The study findings may help experts better predict election outcomes from polling data.
And one thing is clear: Undecided voters should probably not be taken at their word, said Wilson of the University of Virginia.
"They may either have made up their minds, or they are leaning in one direction," he said.
"Perhaps what campaigns should be doing is not persuading people to change their minds but basically getting people who are leaning in one direction out to the polls.
"The problem, of course, is how to figure out who's your voter."
SOURCES AND RELATED WEB SITES