Over at Pollster.com, Doug Rivers has a long post about survey bias and weighting.
We're not talking about standard internet junk polls here, but ones that are actually trying to measure something. This is an increasing problem in the industry: people are much less willing to be surveyed, and the ones who are willing may not be the same as those who are not.
It's a great post if you are interested in this sort of thing, and there have to be at least a dozen people besides me who are. I think Rivers pays too much attention to matching demographics (and not enough to matching behavior), but then that's what he had to work with.