best way to think of multisampling is this.

go in-game and walk on a road, any road. look straigh down (bird's eye view) and check out the nice patterns on the road. notice how they repeat themselves after a while? imagine a square shaped box making up that pattern, with a bunch tiled together to make up the pattern for the road. thats multisample? no not yet.

now instead of looking straigh down on the road change the angle so you can see the road leading off into the horizon. do you see the patterns changing to relate to your perspective of the road, relative to where you are? and if you look futher into the horizon where the road leads... do you still see the same pattern, only smaller? if you do then you've got your multisampling at a super high setting. if you look into the horizon but instead of seeing the same patterns on the ground, you see a muddy, murky blurr that represents the road, then you've got your multisampling on a lower setting. basically your computer will take one pattern and use it over and over and over again to make the terrain. the farther away something is, the blurrier things will get. the higher your settings are, the clearer you will see things at a further distance. with a lower quality videocard, say a geforce 6200le, you can actually see a line where the change will occur from blurr to clear while you're running. better cards will add more transition between what's far and whats near, and will make the world you're in seem seamless. hope that explains it a little.

multisample... take one sample and multiply it across your screen. in the above example, it was a sample of the road.