Good question. It would be a necessary step, so the procedure is very much similar to the the traditional Baron and Kenny (1986) procedure.
If you play around with both models you would probably be able to see that the stats would also change somewhat when you test a 'simple' vs 'mediation' model.
A decent instruction video could be found here: https://www.youtube.com/watch?v=ZI0OlZ6ppBQ
The benefit of AMOS is of course that you can produce model fit indices, and direct and indirect statistics can be readily calculated.