Wow, you might as well ask, "How do you make desert for a meal?", "How do you play the 9th inning of a baseball game?", or "How do you bring a woman to orgasm?" The details completly depend on the meal, the game and the woman.
Nevertheless, I'll tray and give a very general outline to the workflow when I do it. Keep in mind I am not a pro mastering engineer with specialized mastering gear. I am a pro mixing engineer who also happens to do basic mastering when there is not budget or end use need for the Big Boys...
First thing is to take a break between the mixdown and the mastering. I try to make this break at least an hour long, but if the deadline, budget and workload/workflow allow for it, I'd often prefer to let the mixdown marinate in the studio overnight before I go back to master it. The idea is to step away from the console, get out of the room and let both the ears and the brain rest (or at least work on other things) for a while. Coming back to master a mixdown "fresh" instead of chrning right through both the mixing and the mastering in one fell swoop is (for me anyway) a real imperative. The mixdown aways sounds entirely different to me after the break and the work always cones at a higher quality for me that way as well.
After that break the first thing is to give the mixdown a run-through listen and make sure that the mixdown is indeed a keeper. Often times after the above-mentioned break, I'll hear something in the mixdown that got past me when I first made it because I was still too close to the mix. If I need to, I'll then go back to the multitrack, make whatever tweaks I need and make the final mixdown.
Once I am sure I have a keeper mixdown, then I'll look at the waveforms and the levels. It depends totally on the type of content I'm dealing with, if there are dynamics in the mixdown where there are a handful of peaks that are far and above the "average peak" level (so to speak), I'll knock them down manually using the wavefrom editor. Some on this board think I'm an anachronism for using a waveform editor for something as "silly" as actually editing waveforms, but I find this step to be very helpful when it's called for.
The rest of the peaks I might take a limiter to, or I might move right to a compressor; this is a call-em-as-you-see-em thing, it totally depends on the content. Sometimes whether or not to use a limiter or compressor, or which model of each to use can depend as much on the desired sound as it does with technically taming the levels. Having choices here by having more than one plug and/or outboard rack unit is almost as important as having a decently large mic locker is in the tracking phase. I don't have a super-wide selection myself (three limiter and four compressor plugs, plus a couple of channels of outboard limiting and tube and optical compression), but it's enough to matter. What sounds good on a rock anthem won't necessarily sound anywhere near as good on a folk ballad, etc. Having the choice of what to go to is like having the choice of several irons when approaching the green on your favorite golf course. A 5 iron to a fast green 80 yards away is probably just not going to cut it
.
As far as levels; I'm one of the few non-headbangers around here apparently. The stuff I work with rarely, if ever, benefits from crunching a song anywhere near a single-digit RMS, and even the "commercial" stuff that is sounds like crap to my ears when it does. (Listen to Fastball's "The Way" and tell me it would not have sounded a thousand percent better if they had not gone commercial and squashed that otherwise nice song until it was as flat as a pancake.) However, that doesn't mean I won't try and get what I can out of each song level-wise and (more importantly, if making a CD) try to make each seperate song in a group either match in volume or have relative volumes that will evoke the desired results when plays as a group. Typically for me this means I'll wind up with songs somewhere in the -14 to -18dBRMS range, with -16 being far more typical (the real raunch or synth stuff will go to -14 and the highly-acoustic and dynamic stuff might go as low a -18, but those are rare extremes for me.) Others will tell you that even -14 is too quiet and that I'm an a__hole for suggesting otherwise. Take it all as you will.
Once I have the levels set, I might need to apply a little EQ adjustment to the mix, as the tonal balance often changes after compression. This stage is entirely optional, but sometimes a good idea, not only to correct for compression artifacting of the timbre of the song, but also sometimes to help the song fit in with the other songs on a CD project. But the EQ is usually pretty gentle at this stage, If I find that the compression causes me to need more than a few dB of EQ up or down here and there, then that is usually an indication to me that I have either pushed the compression too far, perfromed the limiting and compression process wrongly, or have used the wrong compressor. I those cases I'll go back and perform the compression over a different way.
After I have that all sounding fine, then it's a matter of providing and fades or tail verbs at the end of the song that may be necessary. Again, this depends upon content; some songs have the finishes already arranged into the mix, on others the tail of the song is left ragged for the engineer to take care of.
Still other times the finish is there, but the mastering process itself dictates that the end be re-finished. Often times if there is a slow fade or a noticable verb tail at the end of a song, by the time the song gets through all the processing of mixdown and mastering the finish doesn't sound right anymore, in which case a new fade slope or reverb tail may be applied.
Then I take the resulting product and play it back on "normal" end user systems and see how it sounds, perhaps bringing in another trusted ear for a second opinion, if available. If it sounds passible, I'm done. If not, I'll stick the shotgun in my mouth and pull the trigger with my big toe.
G.