I am attempting to explain a basic principle to you that, frankly, schoolchildren master with little trouble. I could be much more condescending, but for your benefit and perhaps other curious folks, I will explain a bit more.
Let's say there are one thousand people who are absolutely sure how they are going to vote. Because I'm omniscient, I know six hundred of them are going to vote for A, four hundred are going to vote for B. But you don't know that, and you're not sure! You want to know. So you start asking people, but people are busy and you don't have a lot of time. You only get to ask twenty, randomly, from the population.
Now, hopefully, of the twenty, twelve say they are voting for candidate A, and eight say for B. This would be exactly correct (and is technically the most likely outcome), but it is probably not going to happen. In this circumstance, if you were to randomly get 13 for A and then 7 for B...you can see how simple that would be...now we would project our population to be 65 percent for A, and 35 percent for B. And you might even get a crazier result, like 16 versus 4, just by bad luck!
Polls try to sample as many people as they can, and try to get an accurate cross section of the population (if you were to ask only people with landlines, for instance, you would get only older and thus more conservative folk). So because of this uncertainty and, yes, random chance inherent in representative sampling, there is a fairly significant margin of error within every poll.
Now, I feel a little more comfortable in telling you to close your mouth until you understand what I just said.