One stage has access to the individual tracks and the other does not. Every other difference between them — the room, the ears, the tools, what each can rescue — follows from that single fact.
"Can you master my mix" and "can you mix my song" get used almost interchangeably by people outside audio production, and the confusion is fair. Both happen near the end. Both involve processing. Both are trying to make something sound better. But they solve different problems, on different objects, and confusing them is one of the most reliable ways to waste money.
Everything that follows comes from a single difference, which is worth stating before any of the detail: mixing has access to the individual tracks and mastering does not. Every capability and every limitation of each stage is a consequence of that one fact.
Mixing works with the individual tracks — potentially dozens of them, still separate, still editable — and its job is to get them to sit correctly against one another. Levels, panning, EQ so instruments stop competing for the same frequency space, dynamics, effects, automation, all in service of producing one cohesive stereo file.
Every tool covered elsewhere in this path is fundamentally a mixing-stage tool: they are applied to the pieces that make up a song, one at a time, while those pieces still exist as pieces. That is not a rule somebody made up — it is the only stage at which those tools can reach an individual instrument at all.
Mastering begins once mixing is finished. It takes that single balanced stereo file and works on it as one indivisible thing, with no access whatsoever to the instrument tracks underneath. Its job is narrower and different in kind: overall tonal polish, loudness that is consistent within a track and across every track on a release, and getting the material to fit whatever medium it ships on.
That last part is more concrete than it sounds. A CD realistically carries something like 96 dB of usable dynamic range; vinyl closer to 70 dB; cassette around 50 dB across a narrower 50 Hz to 15 kHz window. A mix's full dynamic swing has to be folded down into whatever window the release format actually offers, and it has to still sound deliberate there rather than merely quieter and squashed.
That folding-down is a mastering decision precisely because it depends on the destination — and the destination is frequently not chosen while mixing is happening. You cannot target a window nobody has named yet, and the same mix may need genuinely different answers for vinyl and for streaming.
Mastering can make a good mix better. It cannot rescue a broken one. If the kick sounds right and the bass guitar is wrong, no mastering move fixes that combination, because by then the two of them are in the same two channels — there is no setting that reaches one without reaching the other.
Mastering houses have the stories to match: a client whose mixes were flawed enough that the only honest answer was to send the album back for a remix rather than master around the problem. The release slipped weeks. The alternative was mastering something that could not be saved, and shipping it.
This is also the answer to whether mastering can start before the mix is finished. It cannot usefully, because every mastering decision assumes a fixed relationship between the parts. Change the balance underneath and every move above it is now answering a question that no longer exists.
Two reasons, and they are different in kind.
The first is consistency across a release. An album's songs get mixed across different sessions, different days, sometimes weeks apart, and left alone they drift in level and tone. Mastering is what turns a pile of individually-fine mixes into something that plays as one record and holds up against everything else a listener already owns. Doing that requires having all of them in front of you at once, which is not the state anybody is in while mixing.
The second is ear fatigue and bias, and it is worth being precise about what that means. It is not about hearing damage. It is that whoever has spent hours or days balancing every track has lost the ability to notice the result — familiarity removes exactly the faculty the second stage exists to supply.
Which makes "fresh ears" concrete rather than a figure of speech: a genuinely different room, a genuinely different set of monitors, days or weeks after the mix engineer stopped being able to hear their own work. It is also a different skill in the room. Clients essentially never ask for three decibels at 8 kHz — they ask for a record that sounds big, warm, present, airy, punchy. Turning an adjective into a move is the job.
Described from outside it can sound like a mystery, and it is a short list.
Notice how much of that list is not processing at all. Two of the six are decisions about the release rather than about the sound, which is a large part of why the stage exists as a stage.
What a mastering engineer wants from a mix is more specific than the advice usually given, and one very widely repeated instruction is worth examining.
The instruction is "leave 6 dB of headroom", and the number is not the point. A mix peaking at −6 dBFS that has already been through a bus limiter is a worse starting point than one peaking at −0.5 dBFS that has not. What actually matters is that the dynamics have not already been spent — that the decisions are still available. Peak level is trivially adjustable; range that has been squeezed out is not coming back.
So the useful version is: send a mix that is not already limited, at whatever sensible level it lands, at the session's own sample rate and bit depth, with no dither and no conversion applied on your way out. And send it as you approved it — the mastering engineer needs to hear the thing everybody agreed on, not a variant prepared specially.
Which raises the honest exception: mix bus processing that is genuinely part of the sound. If a mix only sounds like the record with a particular compressor across it, that compressor is part of the mix and should stay. Print an alternative without it as well. That costs one bounce and keeps a decision open that otherwise closes silently.
The sequencing is worth having explicitly, because a surprising number of problems are really scheduling problems.
Recording captures the material. Mixing balances it, and takes anywhere from hours to days per song. Mastering takes the finished mixes and typically handles a whole release in a day or less — which surprises people, and is not a sign that little is happening. It is a sign that the decisions are few, large, and made quickly by somebody who has done it many times, on material that is already right.
Two consequences follow. Mastering is cheap relative to mixing, which is why skipping it to save money is almost always a false economy. And because it is fast, it is the stage most likely to be squeezed by a deadline — with the result that the one stage designed to provide objectivity gets done in a hurry by somebody who has just been told the release is on Friday.
Set side by side, the two lists explain most of the confusion by themselves.
Only mixing can: change the balance between instruments; move something in the stereo field; fix a masking problem between two specific sources; alter the tone of one instrument without altering everything else; edit or replace a performance; change the arrangement by muting something.
Only mastering can: make a set of songs sit together as a release; set final loudness against a platform target; apply the format-specific decisions a destination requires; provide the objectivity of a second room and a second pair of ears; produce the actual deliverable, with its sequencing, spacing and metadata.
Both can: adjust overall tone, and apply dynamics — which is exactly why people conflate them. The difference is not the tool, it is the resolution at which it is applied. An equaliser on a mix bus and an equaliser in mastering are the same device doing incomparable jobs, because one of them is still able to miss the bass guitar and the other is not.
The two jobs are not the same size, and the difference is entirely in the consistency work.
Mastering a single is the narrow version: tonal correction, dynamics, loudness, deliverables. Mastering a record adds the part that only exists across a set — twelve songs mixed weeks apart, each individually fine, that have to arrive as one object. The tonal drift between them has to be reconciled without flattening the differences that were deliberate. The loudness has to be consistent, but a deliberately quiet interlude must stay quieter than the track before it or the arrangement is destroyed. And the gaps and fades have to be decided, which is the most purely musical work in the whole stage.
Which is why an album is not twelve singles and cannot be priced or scheduled as though it were — and why sending tracks for mastering one at a time as they are finished produces a worse record than sending them together, however convenient the first option is.
Most people reading this will master their own mixes at least sometimes, and the honest position is that it is a compromise rather than a sin — but it is worth knowing precisely which part you are giving up.
You are not giving up the tools; a competent chain is affordable and the techniques are learnable. You are giving up the second opinion, which was most of the value. The mastering engineer's contribution is not primarily the equipment — it is that they have never heard this before, they hear it in a room they trust, and they have no investment in the decisions that produced it.
Which suggests how to compensate rather than pretend. Leave real time between finishing the mix and mastering it — days, not hours. Master in a different session, ideally at a different time of day. Use references aggressively, level-matched, because they are the closest thing to an outside opinion you have. And accept that the one thing you cannot manufacture is not knowing what you were going to hear next.
The phrase appears on a great many service listings and it means less than it implies. There is no separate technical process called mastering for streaming; there is ordinary mastering, done with an awareness of what platform normalisation will do to the result.
What it should mean in practice is: master to sit sensibly against the platform's loudness target rather than as loud as physically possible, leave true-peak headroom so the transcode has somewhere to overshoot into, and check the result after upload because that transcode is the only version anybody hears. All of which is covered properly in the delivery course.
What it sometimes means instead is a louder master with a different name, which is the exact thing the normalisation was designed to make pointless. Worth being able to tell the two apart when somebody is selling you one.
Two cases, both real, both worth knowing so the clean distinction above does not become a superstition.
Stem mastering hands the mastering engineer several sub-mixes — drums, bass, vocals, everything else — rather than one stereo file. It restores partial access to the parts, which is exactly why it costs more and why some engineers decline it: it reopens decisions the mix stage was supposed to have closed, and it puts a person who was not in the mix session in the position of re-balancing it.
Mix bus processing is the other direction — mastering-style moves happening during the mix, on the whole stereo signal. Some of that is legitimate and always was. What is genuinely risky is the limiter, because it makes everything sound momentarily more finished and therefore disguises the balance problems the mix stage exists to find. If you are reaching for one mid-mix, the honest question is what it is hiding.
The distinction survives both exceptions, because it was never really about which tools are used. It is about access, and about the fact that a problem between two instruments has to be solved while they are still two instruments.
Mixing has the parts, mastering has the whole. Everything each stage can and cannot do follows from that — which is why a problem between two instruments has to be solved while they are still two instruments.
Students pay for getting unstuck: ask a concept question, routing issue, DAW confusion, or mix decision tied to this lesson.
Free readers can learn from the public Q&A archive. Paid students can ask their own lesson-specific questions and get mentor replies.