A recent study looked at the effects of familiarity on drivers noticing important changes in road signage. As you might expect, when drivers get familiar with a road, they look less and less at the signs because they have committed them to memory. This is basic expert behavior and is generally a good thing.
You might also expect that when the signs are changed, the familiar driver is less likely to notice the change. In this case, they changed a right of way to a yield (making it very important for the driver to slow down/stop because cars coming across the intersection will not be stopping). You might expect that familiar drivers remember the right of way sign so they don't bother looking, and therefore don't notice that it was changed to a yield.
But what this study found is that the drivers do look at the new sign. But they still don't notice the change. My suspicion is that there is so much top-down processing that they still "see" the right of way, even though it now says "yield".
There are several important consequences of this. For one, it means that when you change a design, it is important to take steps to saliently inform users of the change. Kind of like when IVR phone systems have a message that says something like "please listen carefully because the options have changed."
It also makes it important to get your design right the first time. Many web businesses are launching early betas and using the "new" design strategy to launch half-done businesses to get first mover advantage and let customer comments drive future design. I applaud the focus on users, but this study shows that there are consequences of doing this if you let users get too familiar with the early stuff. You may get stuck with some bad features.
Third, it impacts the forensic work that many of my colleagues and I do. When we explain to a jury why a driver blew right through a stop sign that is obvious from the photo of the scene, it is not because the driver was inattentive and irresponsible. It is because he/she was an expert.