Outliers are data that lies far out of the area most of the rest of the data is in.
Sometimes you can explain why the data point. If you were trying to measure the sound from the insects in the trees and a car horn honked, you could reasonably exclude that data as an outlier. But you might also want to exclude a data point that is 10 Standard deviations from your mean. Just don't delete so many points it invalidates what you are doing.
I'll quit here, as I think I have answered your question. Ask again if it is still unclear.
Also see:
Outliers Math Forum - Ask Dr. Math
Linear Mode
