Actually, all durations, until they've occured, are estimates. In
general, when you put in an estimate, it's just your best guess,
preferably, in my experience, on the pessimistic side.
Another option, though now know to be statistically unsound, is called
PERT Analysis. In this methodology, you make 3 estimates of each tasks
duration, optimitic, pessimistic, and most likely. You plug them into a
standard formula, which takes a weighted average, giving the most
likely estimate 4 times the value of the other 2, and used that average
as the duration estimate. In Project, you can implement this by turning
on the Analysis toolbar, then clicking on the PERT Analysis button,
which brings up the PERT Analysis toolbar. From there, you can enter
the 3 estimates either in a form or a table. When you click on the
"Calculate PERT" button, and it will do the math and enter the result
into the duration field.
I can't stress too strongly, that this method is terribly outdated and
misrepresentative of the real implications of multiple potential
duration estimates. The statistically correct way to do this is to use
Monte Carlo analysis, and there is software out there that does it. It
gives far more meaningful results than PERT Analysis. I'm not enought
of a statistician to explain it all, but have read enough to be
convinced that it is so.
Best of luck, however you proceed.