Im my regression class that I am teaching, we recently got to the topic of variable selection.  We covered the standard forward, backwards, and stepwise selection procedures, and I then went out of my way to caution them about using these procedures.  In trying to explain this to them, I believe that I have come up with a fantastic analogy for explaining this to students:

Automated variable selection techniques are like putting together the border of a puzzle:  It’s a great place to start, but you’ve still got a long way to go to be finished.  -StatsInTheWild

Cheers.