There are basically three ways I'll use compression in my mixes
I do mostly television mixes; and, as I've mentioned in another post, dialogue has to sit in a specific range within the mix. I will get the overall track very close with just volume automation, then use some gentle compression to smooth it out and get it to sit in the correct range. I'll use a bit more if it's an element like narration, where we never show the owner of the voice, just to establish it as something outside of the context of the image itself.
Occassionally, I'll use compression to help lift a single element to the front of the mix. I usually do this when I like the overall dynamics and balance of the mix but feel that something needs just a little more presence. Again here, it's just a very gentle compression to add a little weight to that element.
And lastly, I use a limiter on the master output bus. In television, nothing is supposed to peak above -10db full scale. I try to control that in the mix, but there's only so much time. So, I use that limiter only to catch any transients that escaped my attention. I set it to catch those transients and have little to no affect on the rest of the mix. It may be changing now with the full digital switchover here in the U.S., but it used to be a systems issue for television broadcast equipment. They couldn't handle a signal that was hotter than that, and it would cause all kinds of problems. Of course, this is something I've heard and never bothered to get confirmation on. All I know is programs often get kicked back by the networks if you have anything peaking over -10dB. As far as I understand it, many other countries have the same rule (we sell a lot of our programming overseas as well).