I've got an idea that may make interpreting poll numbers a little easier. Take out the undecided voters. As an example, the Pew poll (Oct 16-19 national) showed Bush's approval rating at 44%, while 48% disapprove. Without the undecideds (8%), the approval numbers are 48%, leaving a majority of 52% in disapproval of the president.
If the election were held today, it is possible that Kerry could win with 52% of the vote.
Getting rid of the undecided voters allows for a clearer picture of what the final vote tally will look like on election night, because "undecided" will not receive 6-10% of the vote on election night, like it does in most polls. When the final results come in, Kerry and Bush are likely to split 98-99% of the vote between themselves, with the remaining 1-2% split among Greens, Libs, or other third parties. To get a better estimate of what the actual results might be, throwing out the undecideds makes sense.
It has been said elsewhere (multiple elsewheres, so no link - sorry) that at least 60% of undecided voters are likely to vote for the challenger, and no more than 40% are likely to vote for the incumbent. This year, however, the remaining undecideds appear (anecdotally) to dislike both candidates enough that many may either stay home or vote for someone other than Bush or Kerry.
Taking out the undecided voters is really just a way to make the poll numbers a little more intuitive. You can see who is ahead at a glance, and the results conform to the likely results on election night. As with any poll, it may make sense to give Kerry at least one or two points extra to counter weight against the Bush bias of many polls and the massive Democratic GOTV campaign that is currently underway.