Here is some basic psychology for those interested. It concerns memory and how it can be used for good design. It is already widely used in marketing and may be useful.
The serial position effect concerns how well something is remembered. It came from experiments where participants were given a list of things to remember and asked to recall them as best as they could. Numerous studies found that items at the start and at the end of the list were recalled best: items in the middle were recalled less well.
Why? The first-presented items may be processed to a greater degree. There are many explanations for this, but a useful one is that later items are similar and so are processed less because the mind prefers novel stimuli - another reason for my website's layout being different. Later items are also closer to recall but this is a poor explanation - quite possible, with repetitive stimuli, later items are encoded using elements from previous items (much like video compression) which causes a degree of the earlier items being "written over". Later items are not written over because similar stimuli are not presented for them to be written over with. Anyway, this is one explanation and there are many out there. The important thing to remember is that the earliest and latest items are remembered best.
What use is this to interface design?
Quite often, people are asked to remember things. For example, if using SPSS to analyse statistical data, I would have to remember the design of the experiment and the nature of each variable (how many levels, what are the levels, is the variable within- or between-subjects, the ANOVA model, etc). SPSS helpfully puts up a dialog but I have to remember all this information and input it. The serial position effect says that the first and last information I encounter will be remembered best so the most important parts need to be placed there. With some other tasks, I need some information and can work out others in the dialog. In this case, it is best to present the essential information first and last, and the information that can be worked out in the middle. Using the SPSS example above, I can work out the levels of the variable in the dialog itself, so those can be left in the middle.
Doing this reduces the demands of the task. This means that it is less likely to encourage errors or even "rollbacks", when a user has to quit the dialog and go back to find the information (often writing it to a piece of paper to aid recall).
btw, I dislike dialogs enormously. That's another article.