Gallup is almost always right, except for Dewey-Truman and Ford-Carter they’ve picked the winner:
http://www.gallup.com/poll/9442/election-polls-accuracy-record-presidential-elections.aspx
In 2004 Gallup predicted a dead even race, 49-49. Bush prevailed by 2.4 points.
Their poll of decided voters actually showed a 2 percent gap 49-47 for Bush. But then they went ahead and assigned 90% of the undecided voters to Kerry on the theory that the undecided always favor the challenger. In fact, the undecided broke mildly for Bush.
How did they treat the undecided now? Anyone knows?