Careful readers of this space may recall this from me:

Quote Originally Posted by Jeff Dieffenbach
A huge political pet peeve of mine: national polls showing Presidential Candidate X against Presidential Candidate Y. Obama 46, McCain 44, that sort of thing.

Those polls are meaningless in the general election. Rather, if you want to report something meaningful (although even this is debatable given the recent validity of polls), it makes more sense to project electoral votes using state by state polls. Fortunately, www.electoral-vote.com does exactly that.
For the 2004 election, I'd done a similar analysis for Kerry v. Bush. I was planning to do the same for the current race, but found that Five Thirty Eight (whom I'd also examined in 2004 and 2008; and yes, I'm aware of Rasmussen, but no, I don't give it as much credence) now presents the race in almost exactly the way that I'd been thinking. Rather than reinvent their wheel, I'll simply link to their work.

The right column of the Five Thirty Eight blog shows electoral vote projections over time, starting in June and ranging through the present. Obama's lead hovered in the 290 to 250 range for the first two months, then began to widen to its current margin of roughly 320 to 220.

For Romney to close this gap, he'll need to steal on the order of 50 electoral votes (EVs) from Obama's column. Five Thirty Eight gives Romney the best chance of doing so in FL (29 EVs, 30% of winning), CO (9 EVs, 25%), IA (6 EVs, 25%), and VA (13 EVs, 25%). What happens if Romney pulls off this unlikely quadfecta? He'd steal 57 Electoral Votes, bringing his total to 275 and eeking out a victory. What odds does Five Thirty Eight put on Romney pulling off this or a similar coup? About 16%.