The short version
Mixing is about the relationship between the parts: how loud the vocal is against the snare, where the guitars sit, whether the bass and the kick are fighting. Mastering starts after that is settled, treats the whole mix as a single piece of audio, and asks a different question — does this hold up everywhere it will be heard, and does it sit properly beside other records?
The clearest test of which one you need: if the problem can be described as “this element is wrong”, it is a mix problem. If it can only be described as “the whole thing is wrong”, it might be a mastering problem — and it is usually still a mix problem.
They are done in that order, by different people more often than not, on different material: a mix engineer works from every recorded track, a mastering engineer from the single stereo file the mix produced. Everything else about the two follows from that.
What mixing does
A mix engineer has every element separately and can change any of them. Level, panning, equalisation, compression, reverb and delay, automation over time. Most of the work is making room: two instruments occupying the same frequency range and the same moment will mask each other no matter how loud either one is, and the fix is almost never “turn it up”.
The moves a mix is made of, roughly in the order they tend to happen:
- Balance — the level of each part against the others, which is most of the mix and the part most often skipped in a hurry to reach the plug-ins.
- Placement — panning and depth, so that parts which share a frequency range at least do not share a position.
- Corrective EQ — removing what masks something else: the low-mid build-up in a guitar that is hiding the vocal, the boxiness in a snare. The vocal case is the one worth learning first, because most of the fix happens on the other tracks.
- Dynamics — compression on individual parts so that a vocal stays audible through the whole phrase, and a bass stays consistent under the kick.
- Space and movement — reverb, delay, and automation that changes any of the above over time, because a static mix of a song that develops is a mix of its first verse.
That is why mixing decisions are hard to undo later. Once a mix is bounced to a stereo file, the vocal and the snare are the same numbers, and no amount of processing separates them cleanly again. Separation gets closer than it used to, and it is still an estimate rather than a recovery — what it can and cannot do is worth knowing before you rely on it as an undo button.
What mastering does
A mastering engineer has one stereo file and a much smaller set of moves: broad corrective equalisation, gentle multiband dynamics, stereo width, and a limiter setting the final level. The restraint is the discipline — every move touches everything at once, so a 2 dB boost to fix a dull top end also brightens the cymbals you were happy with.
The other half of the job is preparation rather than sound: making sure the track holds up on a phone speaker and in a car — where what survives a fold to mono is the thing most often missed — that it does not clip when a streaming service re-encodes it, that its level is sensible relative to the rest of the record and the rest of the world, and that the file itself is correct: sample rate, bit depth, fades, spacing between tracks on a release, and the metadata a distributor asks for.
There is also a job that is not processing at all: a second, fresh pair of ears on a record the mixer has heard four hundred times. A good deal of what a mastering engineer sends back is not a master but a note — “the vocal drops out in the second chorus” — that gets fixed in the mix. That is the system working.
Mixing vs mastering vs producing
Production is the stage before either: deciding what the song is, what plays, how it is arranged and how it is recorded. A producer may mix and may not; in a great deal of modern music the producer, the writer and the mix engineer are the same person at the same laptop, which is exactly why the boundaries feel blurred.
The order still holds even when one person does all three. Arrangement decisions belong to production, because a part that should not be there is cheaper to cut than to mix around. Balance decisions belong to the mix. And loudness and delivery belong to mastering, last — which is the one stage that most benefits from being done by someone other than the person who made the other two sets of decisions.
What to hand a mastering engineer
Whoever or whatever is doing the mastering, it wants the same thing: the finished mix, with nothing done to it that mastering would do again.
- No limiter on the mix bus. A mix already pushed to the ceiling has had the mastering decisions made for it, badly; the engineer’s first job becomes undoing them. A few decibels of headroom — peaks around −6 dBFS — costs nothing.
- The session’s own sample rate and bit depth. Whatever it was recorded and mixed at, delivered at 24-bit or 32-bit float. Converting or dithering before mastering is a job done twice.
- No fades, or fades you are sure of. A fade is easy to add and impossible to remove; leave the tail and say where you want it.
- A reference or two, and a note. Records in the genre you think sound right, and a sentence about what you are worried about. A mastering engineer working blind will make a good record; one who knows the vocal was the fight will make yours.
Where loudness fits
Mastering is where a track’s final loudness is decided, which used to matter enormously and now matters differently. Streaming platforms normalise playback by perceived loudness, so a master that is 3 LU louder than another simply gets turned down by 3 dB on the way out. What survives normalisation is everything else the loudness cost you: the dynamics you compressed away are gone whether or not anyone hears the volume.
The one thing worth being strict about is inter-sample peak. A master that reaches above the platform’s ceiling between samples distorts in their encode, after normalisation, where nobody hears it until it is public. Both are covered in what LUFS actually measures, the per-service figures are in the streaming targets guide, and a finished master can be checked against them with the loudness checker before it goes anywhere.
Can mastering fix a mix?
A little, and less than people hope. A master can correct a broad tonal tilt — a mix that is uniformly dull or uniformly thin — because that is one problem affecting everything. It cannot bring up a buried vocal, because the vocal is not a separate thing any more.
The practical rule: if you find yourself hoping mastering will fix something specific, fix it in the mix. If you are asking for a general polish and a sensible level, that is what mastering is for.
The exception that is worth knowing about is stem mastering, where the engineer is handed a handful of grouped stems — drums, bass, vocals, everything else — rather than one file. It gives mastering a little of the mix’s reach, at the cost of a little of its objectivity, and it is a reasonable answer when a mix is close but one group is off. If the session is gone and only the bounce survives, separation can produce those groups, with the caveat that they are estimates of the stems rather than the stems.
Doing both yourself
Plenty of records are mixed and mastered by one person, and the reason it is harder than it looks is not the tools. It is that the two jobs need different ears, and you only have one set. A mixer is inside the song, listening for the parts; a mastering engineer is outside it, listening for the record.
Four habits get most of the way to a second person:
- Separate sessions. Bounce the mix, close the project, and master the bounce in a fresh session — ideally on a different day. If you can still reach for the vocal fader, you are still mixing.
- References at matched level. Compare against released records with the levels matched, because a louder version wins every comparison regardless of what it did.
- Level-matched bypass. Every mastering move should be judged with the processing bypassed and the gain matched. A move that only sounds better because it is louder is not a move.
- Measure, then decide. The numbers are not the master, but they catch what tired ears miss — a true peak over the ceiling most of all.
The step-by-step guide to mastering a song walks the second half of that in detail.
Where AI mixing and mastering fit
The same division applies to automated tools, and it is the first thing to check when someone offers “AI mixing and mastering” as one button. A mixing tool works from the separate parts; a mastering tool works from the finished stereo file. A tool that only accepts a stereo file is a mastering tool, whatever it is called, and it cannot do anything to your vocal that a mastering engineer could not.
Euphona keeps the two apart for exactly that reason. AI mixing takes stems and produces a balanced starting mix — masking-aware levels, corrective EQ, bus processing — with every move listed, so it can be read, kept, or overridden. AI mastering takes the finished mix, sets a loudness and tonal target for the material rather than for a platform, and reports the chain it applied. Both work from one analysis of the track, so the mastering stage knows what the mixing stage did; neither pretends the other stage’s decisions can be made from its side of the line.
Keep reading