Before you start: is the mix finished?
Mastering works on one stereo file, which means every decision it makes applies to everything at once. That is the whole constraint, and it decides what mastering can and cannot do for you. If the vocal is too quiet, mastering cannot fix it — the vocal is not a separate thing any more, short of separating it back out, which is a different job with its own costs. If the whole record is a bit dull, mastering can absolutely fix that, because dullness is one problem affecting everything. Mixing versus mastering draws the line in full.
So the test before you begin: can you name a specific element that is wrong? If yes, go back to the mix. If the only complaint you can articulate is about the whole thing at once — it is quiet, it is dark, it does not sound like a record — you are in the right place.
Bounce the mix without a limiter on the master bus and with a few decibels of headroom. Peaks somewhere around −6 dBFS is plenty. This is not because headroom is magic; it is because a mix already squashed to the ceiling has had the decisions made for you, and mastering it means undoing someone’s work first.
Step one: listen and measure
Play the track through once without touching anything. Then measure it — our free loudness checker does it in the browser without uploading the file — because your ears have opinions about level that instruments do not.
The three numbers worth knowing before you start:
- Integrated loudness (LUFS) — how loud the whole track is, perceptually. This is the number streaming services use to decide how loud to play you, and what LUFS actually measures covers how it is arrived at.
- True peak (dBTP) — the highest the waveform actually gets, including between samples. Not the same as your DAW’s peak meter, and the difference is where clipping in encoders comes from. If the file is already distorting rather than merely peaking, the clipping detector finds where.
- Loudness range (LRA) — how much the loudness varies across the track. A folk record and a techno record should not have the same number here, and a number near zero means you have compressed the life out of it.
Then pick two or three reference tracks — commercially released records in the same genre that you think sound right — and measure those too. You are not trying to match them exactly. You are calibrating your sense of where the material you are working on sits.
Step two: fix the tonal balance
Tonal balance is the first move because everything after it responds to it. A compressor reacts to what you feed it, so equalising after compression means the compressor was reacting to a spectrum you have since changed.
Work in broad strokes. Mastering equalisation is measured in fractions of a decibel to a couple of decibels, over wide bands. If you find yourself reaching for a 6 dB cut with a narrow Q, you have found a mix problem and you should go and fix it in the mix.
The moves that come up over and over:
- A gentle high-pass below about 30 Hz, to remove subsonic energy that costs headroom and that nobody will hear.
- A decibel or two of low-mid reduction somewhere between 200 and 400 Hz, where home mixes most often accumulate mud. If it is the vocal specifically that is buried there, that is a mix move — how to EQ vocals covers it.
- A wide, small high-shelf lift above 8 kHz if the record needs air — but check it on headphones, because a shelf that sounds like clarity on speakers can sound like sibilance in someone’s ears.
Compare with the equaliser bypassed, and match the levels when you do. A boost sounds better than a bypass almost regardless of what it did, because it is louder. That is the single most reliable way to fool yourself in this entire process.
Step three: dynamics, carefully
Compression at the mastering stage is about consistency, not about impact. You are gently reducing the difference between the loudest and quietest parts so the record holds together — a decibel or two of gain reduction at a slow ratio, most of the time.
Multiband compression is the more surgical version: it lets you control the bass without the vocal triggering it, which is useful when a track has a low end that moves around. It is also the fastest way to make a record sound processed, because independent bands can drift out of relationship with each other. Use as few bands as solve the problem.
The honest default is: if you cannot hear what the compressor is doing when you bypass it at matched level, it is probably doing about the right amount.
Step four: loudness and the limiter
Here is the part that has changed most, and that most advice on the internet has not caught up with. Mastering louder does not make you louder. Every major streaming platform normalises playback by measured loudness. Master 3 LU above the target and the platform turns you down 3 dB. The listener hears the same volume; you have simply given away 3 dB of dynamic range for nothing.
So set the limiter’s output ceiling first — −1 dBTP is the safe default — and then bring the input up until the integrated loudness lands where you want it. Around −14 LUFS suits most streaming music; the per-service targets are published and worth checking rather than remembering. Louder is a legitimate artistic choice for some genres — mastering hip-hop works through one where it usually is — but make it a choice rather than a reflex.
Why −1 dBTP and not 0? Because lossy encoders reconstruct a waveform that can overshoot the samples it was built from. A file that peaks at exactly 0 dBFS can clip after encoding, in the version everyone actually hears, in a way you will never hear locally.
One current wrinkle worth knowing: Spotify’s documentation asks that masters louder than −14 LUFS keep their true peak under −2 dBTP rather than −1, because they are about to be turned down and re-encoded, and the extra decibel is the margin for that. The Spotify page and the Apple Music one go through each service’s specifics, including why one master still serves both.
Watch the gain reduction meter. Consistent limiting of a decibel or two is normal. Six decibels of reduction on every chorus means the limiter is doing the compressor’s job and doing it worse.
Step five: check it somewhere else
The mastering engineer’s real advantage over you is not their converters. It is that they have heard a thousand records in their room and know exactly what it lies about. You can substitute for that with variety.
- A phone speaker. This is where the midrange either survives or does not.
- Cheap earbuds — what most listening actually happens on.
- A car, if you have one. Cars have terrible, revealing low end.
- Quietly. Balance problems hide at volume and stand out at conversation level.
Then leave it overnight and listen once in the morning before you commit. Ear fatigue is real and it consistently pushes people toward brighter and louder.
Step six: make the actual files
Deliver a WAV or FLAC at the sample rate you worked at, 24-bit, with no dither if you are staying at 24-bit. If you are producing a 16-bit file for CD, apply dither once, at the very end, and never twice.
Do not upload an MP3 to a distributor. Every platform makes its own lossy version from what you give it, and giving them a lossy source means they encode an encode.
Check the top and tail: no clicks at the start, a clean fade at the end, and silence that is actually silence rather than the room. If it is part of a release, check the gaps and the relative levels between tracks — an album where track four is 2 LU quieter than everything around it will be noticed even by people who cannot say why.
The four mistakes that matter
- Comparing at different volumes. Louder always wins. If you take one thing from this article, make it this: match the levels before you compare anything, at every stage, without exception.
- Mastering to be loud. Normalisation removed the benefit and kept the cost. The loudness war ended and the terms of surrender were favourable.
- Fixing mix problems here. If you can name the element, go back. Every move at this stage touches everything, which is precisely why the mix stage is where a named problem gets solved.
- Trusting one room. Your monitoring lies to you in a specific, consistent way, and the only cure is hearing the track somewhere that lies differently.
Keep reading