So as you all know, I have been a bit obsessed with politics over the last few months (its gotten to the point that after the election happens I am not sure what I am going to do with myself), and during the last week or two I have become more and more obsessed with polling. Now being a scientist, I am fully aware that polling, especially polling averages, are susceptible to bias and human error the same way that all data are ... so I generally look at two or three different polling sites, these days they are Pollster.com, RCP and fivethirtyeight.com. These days, looking at any of these sites is satisfying, as they all show a significant lead for Obama, looking something like this one, taken from RCP (this shows what the election would look like today with all states voting as they are currently leaning, according to the latest polling) ...
It is important to keep in mind that poll results, especially averages, can be altered by which polls to include, sampling likely vs. registered voters, sampling disproportionately high numbers of a particular group (i.e too many democrats or white women or hispanics or any other group that trends toward a particular party or candidate). You could also take more or less polls at a particular time, resulting in a disproportionate polling average bump for one or another politican (I read a report that this has actually been happening, there are far more polls taken when McCain is up than when Obama is up, although now he has been up so consistently that that is changing ... but I cannot find the page linking to this idea).
With all that said, it is pretty easy to figure out a poll or polling average bias, you can compare it to other polls taken at the same time and find the outliers or see how polls differ. In this regard, fivethirtyeight.com just published a really interesting article on this effect ... if you are interested in polling at all it is worth reading.
So I generally look at two or three different polling sites, these days they are Pollster.com, RCP and fivethirtyeight.com.