...who may or may not have voted yet, but that help the MSM make the polls "show" a horse race by excluding them.

Unlikely voters are people who are registered but don't really consider it a high priority to actually go to the polls to vote. That is why GOTV efforts are huge. To convince them how important it is.

I was told it was based on various demographic factors and how you answer their questions. I think it is determined by each pollster. It has never sounded legitimate to me.
Maybe someone has more detailed knowledge.

If it reveals that you vote in just about every election, you are a likely voter. If you've been registered for ten years but you've only voted once or twice in all that time, you're not a likely voter.

...I personally think they're taking the number of registered voters and multiplying by a certain percentage to get as low a number of Likely Voters for the Democrats as possible. In the meantime, they're using a much higher percentage to arrive at the number of GOP Likely Voters.

For instance, a pollster calls 1000 Republican and Democratic Party Registered Voters. The pollster multiplies the Democratic Party RVs by 90% getting 900 (45%) Likely Voters (LVs) because, according to the pollsters, "everybody knows" Dems are less enthusiastic about voting. At the same time, they use a multiple of 95% to arrive at the number of GOP LVs which equals 950 (47.5%).

That allows the pollsters to present the following poll:

Dem LVs = 45%
GOP LVs = 47.5%

The same database that identifies registered voters also contains information about how often those voters actually cast a ballot. There's no need to use some algorithm to create a sample of "likely" out of a larger pool of "registered" if you already have the actual information available.

Another major question pollsters ask is if they voted in 2010. If they didn't then they are not considered likely voters. A lof of Democrats especially minorities did not vote in 2010. This can only be good news for the Obama team.

They ask me how likely I am to vote, on a scale ranging from "very likely" down to "very unlikely." I assume anyone who says likely, somewhat likely or very likely, is called a likely voter.

They count on the respondent to give a factual answer, just as they count on the rest of the answers to be factual.

If someone answers no or says that they are not registered to voter, the poll or likely voters would terminate the call at that point. A poll using all respondents would often include many people who will not vote in the election, even if they do have opinions on who they prefer.

Pollsters massage the numbers all the time to make sure their polls are in-line with what they expect or want them to say. It would be interesting to see the demographic changes compared to the results over time for each pollster.

Gallup, you have voted in the last two elections... period. This excludes those who voted in the Presidential in 2008 that will also vote in 2012... as well as newly registered voters and those who just voted in the Primaries for the first time.

Some polling companies are a tad more generous, but not much.

This is why LV models also tend to oversample the South, which has a heavy retired population that TENDS TO VOTE.

See #10 below.