The brain has its work cut out for it when it comes to interpreting visual sensory input. It must render a three-dimensional representation of the world out of the two-dimensional image that falls upon the retina. Often this 2D image can be ambiguous; that's where top-down processing comes in. Top-down processing is when your perception is influenced by prior experience and expectations. (It's the counterpart to bottom-up processing, wherein your perception is influenced directly by fresh sensory data.) As a nice example of a top-down effect, take a look at this picture of a Martian crater:
To most people, the image on the left looks like a dome, and the one on the right looks like a crater. In fact, it's the same image, turned 180 degrees. The image is two-dimensional. To impose concavity upon the image, your brain makes use of what it knows about lighting. On Earth, objects tend to be lit from above, so for the image on the left, your brain assumes that it's looking at a dome lit from above rather than a crater lit from below. That's top-down processing; your preconceived notions of how light falls on objects affects how you perceive an ambiguous shape.
One of my favorite illusions resulting from misdirected top-down processing is the inverted mask illusion. Here, your assumptions about light and the shape of faces (noses tend to stick out at you, not in away from you) lead your brain to create an incorrect 3D model of what it's seeing, and from that incorrect model your brain perceives motion where there is none.
One more illusion: click here to see a spinning silhouette.
Most people tend to see this figure as spinning clockwise, her right leg extended. But with a little practice, you can train yourself to flip her mentally and see her spinning counterclockwise with her left leg extended (apparently looking at her shadow may help you make the switch). Since it's a silhouette, you have no definite information as to which leg is in front of the other, or whether the extended leg is pointing towards you or away from you. But once your brain picks a particular model, it interprets motion based on that model.
It just goes to show that sometimes (for perceptual psychology in particular) one of the best ways to see how a system works is to look at how and when the system goes wrong.
(Thanks to Bad Astronomy for the first and third demos)