The actual difference
Mixing is the stage where the individual tracks of a song are balanced against each other: levels, EQ, compression, panning, effects. You have access to every element separately, so you can change the relationship between any two of them.
Mastering takes the finished stereo mix and treats it as one unit, preparing it for distribution: final loudness, tonal balance, true peak headroom, and consistency against other releases. You no longer have the parts, only their sum.
That difference in access is the whole story. A mastering engineer working on a stereo file cannot turn the vocal up without turning up everything sharing its frequency range at that moment. This single constraint decides which problems each stage can solve, and it is why the order matters so much.
Symptom by symptom: which stage is at fault
A useful test: does the problem exist between two elements, or does it apply to everything at once? The first is a mix problem, the second belongs to mastering.
- The vocal disappears in the chorus. Mix. Something is masking it, and only separate access can move one without the other.
- The whole track sounds dull. Mastering, usually. A broad tonal tilt applies to everything equally.
- The kick and bass blur together. Mix. Two elements competing for the same space.
- The track is quieter than other releases. Mastering. This is a level and loudness decision.
- The mix sounds muddy. Mix. Accumulation between instruments in the low mids.
- Peaks distort after uploading. Mastering. A true peak ceiling that was set too close to zero.
- It falls apart in mono. Mix, most of the time. Phase relationships between tracks.
- It sounds squashed and tiring. Mastering, if the master compression did it. Mix, if the individual tracks were already over-compressed.
Problems only the mix can fix
These share one trait: they involve the relationship between two or more elements, which no longer exists as a separate thing once the file is bounced.
- Masking. Two sounds occupying the same frequency range at the same moment.
- Muddiness. Accumulated energy in the low mids from many sources at once.
- Balance. An instrument that is simply too loud or too quiet relative to the rest.
- Phase between tracks. Polarity and alignment issues between individual recordings.
- Arrangement congestion. Too many parts in the same octave, which EQ can only disguise.
A mastering engineer facing any of these will do what they can with broad strokes and will usually tell you the same thing: this needs to go back to the mix.
Problems only mastering can fix
These apply to the file as a whole, which is exactly what mastering has access to.
- Loudness. Setting integrated LUFS to the target for the intended platform.
- True peak headroom. The limiter ceiling that keeps lossy encoding from distorting.
- Overall tonal tilt. A broad correction when the whole file leans dark or bright.
- Consistency across an album. Making several tracks sit together as a body of work.
- Delivery formats. Separate versions for streaming, social media and club playback.
Notice how few there are. Mastering is a narrow stage with a narrow set of powers, and that is by design: by the time you reach it, the creative decisions are supposed to be finished.
Why mastering cannot save a broken mix
The phrase “we will fix it in the master” is the mixing equivalent of “we will fix it in post”, and it fails for the same structural reason: the information needed to fix it is no longer separable.
Consider a vocal buried under a rhythm guitar. In the mix you lower the guitar by 2 dB around 3 kHz and the vocal appears. In mastering you can boost 3 kHz on the whole file, which raises the vocal and the guitar together, so the balance between them does not change. You have made the track brighter and harsher without solving anything.
Worse, mastering amplifies existing faults. Compression and limiting reduce the difference between loud and quiet parts, which brings up whatever was sitting in the background: room noise, a resonance you never noticed, the tail of a reverb that was slightly too long. A clean mix gets tighter. A muddy mix gets louder and muddier.
Diagnosing it with measurements
You can settle the question before spending money on either stage, because the two categories leave different measurable signatures.
Whole-file values point at mastering. Integrated loudness far from the target, true peak above the safe ceiling, a loudness range collapsed under 3 LU: all of these describe the file as a unit, and all of them are mastering decisions.
Relationships point at the mix. Spectral balance that departs from the reference shape for your genre in a specific band, phase correlation dipping negative in one section but not another, energy piling up between 200 and 500 Hz: these describe how elements sit against each other, and no mastering move separates them again.
These are computed values rather than opinions, which is what makes them useful for this particular question. The same file always produces the same numbers, so the diagnosis does not depend on whose ears are listening or how tired they are.