Can you tell the difference between lossy and lossless?

Many people on DJ and producer discords and on reddit argue about this all the time. One person says he can clearly hear the difference between a lossless WAV and a 320kbps mp3 on his focals and the other says there is no way you can hear a difference as very little above the 20kHz is cut and the human ear can’t hear those frequency anyways . The honest answer is somewhere in between — and it depends on a few things: the bitrate, the codec, the material, the gear, the room, and — most of all — whether the test is blind.

What follows is the evidence as it actually stands, without the audiophile mysticism and without the smug "it's all placebo" counter-mysticism. Both camps are right about something. Both are wrong about more than they admit.

What the science says

The gold standard for testing this is the ABX test: you're played sample A, sample B, and then X, which is secretly one of the two. You just have to say which. No labels, no expectations, no peeking at filenames. It's brutal, and it's humbling.

The setup matters as much as the concept. A proper ABX rig level-matches the samples to within a fraction of a decibel — because humans reliably prefer whichever version is slightly louder, and a 0.5 dB advantage will masquerade as a "quality" difference in naive comparisons. It also hides everything except the sound: no format badges, no price tags, no waveform displays. What survives that procedure is an actual audible difference. What doesn't was something else.

Decades of controlled testing have shown a pretty consistent picture. At low bitrates — 96 or 128 kbps — most people can pick out the MP3 fairly reliably. At 320 kbps MP3 or 256 kbps AAC, the vast majority of listeners, including trained audio engineers, score statistically indistinguishable from guessing. Modern lossy codecs at high bitrates are genuinely excellent at their job.

The classic public demonstration was a series of community listening tests run through the 2000s on the Hydrogenaudio forums, where volunteers ABX'd codecs at various bitrates on killer samples. The pattern that emerged — transparency clustering around 128–192 kbps for good encoders on most material, with stubborn exceptions on a handful of pathological sounds — still describes the landscape today, because the psychoacoustic model the encoders exploit hasn't changed: your ears haven't had an upgrade since.

When the difference IS audible

That said, there are real cases where lossy encoding shows its seams:

• Low bitrates. Anything at 128 kbps or below usually has audible artifacts — a "swishy" quality on cymbals, pre-echo on sharp transients like castanets or harpsichord, and a general dullness up top.
• Killer samples. Certain sounds are notoriously hard to encode: harpsichord, glockenspiel, applause, castanets, dense noisy passages. Even good codecs can stumble on these.
• Transcoding. An MP3 made from another MP3 degrades fast. Generation loss is very audible — each generation reinvents the artifacts of the last, and streaming a song that was ripped from YouTube and re-uploaded is how a track ends up sounding like it's underwater. If you've ever wondered why the same song sounds worse on one platform's user uploads than everywhere else, this is usually why.
• Critical listening on good gear. In a quiet room, on revealing headphones or speakers, with music you know intimately — some people really can pick out high-bitrate lossy files more often than chance.

That last bullet deserves its honest asterisk. The people who beat chance in controlled tests are real, but they're rare, they tend to be trained listeners who know exactly what artifact to listen for, and their margins are thin — we're talking 60–70% correct identification where 50% is guessing, not a night-and-day revelation. If someone describes 320 kbps MP3 versus FLAC as "obvious" or "like a veil lifting," they're describing something the controlled literature says essentially doesn't happen. 

The myths that won't die

A few claims circulate so persistently they deserve direct answers:

"Lossless has more air and space." What people usually mean by "air" is content above 15 kHz. On a recording that has some, a low-bitrate file genuinely removes it — but at 256 kbps and up, that band is preserved to a degree blind tests can't separate from the original. The "air" was in the mastering, not the format.

"You can feel the difference even if you can't hear it." Controlled tests measuring exactly this — with listeners describing their experience unaware of which file played — find nothing beyond chance. The feeling is real; its cause isn't the codec.

"Hi-res (24/96) sounds better than CD quality." This one collapses for a different reason: when hi-res releases do sound better, it's because they're different masters — more dynamic range, less limiting — not because of the sample rate. Compare the same master at 16/44.1 and 24/96, level-matched and blind, and the difference disappears. The music industry often sells you the better master only in the fancier wrapper, which is a marketing decision, not physics. (Ultrasonic content above 22 kHz is inaudible by definition; the case for 24-bit depth is about production headroom, not playback.)

"Expensive gear reveals the difference." Revealing gear reveals artifacts that exist — that's true, and it's why killer-sample testing uses good headphones. But it cannot reveal information that was never discarded. Better equipment shrinks the plausible excuses for hearing a difference; it doesn't create the difference.

Why the codecs are so hard to beat

It helps to know what a modern encoder is actually doing, because it's not "deleting detail" in any naive sense. It's exploiting two well-mapped quirks of human hearing. The first is the absolute threshold: quiet sounds below a certain level simply don't register, so encoding them is optional. The second — the big one — is masking: a loud sound makes nearby frequencies, and nearby moments in time, temporarily inaudible. A cymbal crash doesn't just get heard; it deafens you to subtler content around it for a few milliseconds.

Lossy encoders maintain a running psychoacoustic model of what's masked at every instant, and spend their bits only on what the model says you'll actually perceive. Done well, the information they discard is information you were never going to receive anyway. That's why the ABX results are what they are: at high bitrates the encoder's model of your hearing is more accurate than your confidence in your hearing.

Where the model strains is exactly where the killer samples live: sharp transients in quiet passages (pre-echo leaks backward past the masking window), sustained noisy textures like applause (too much unmaskable information to fit the budget), and artificial electronic sounds the psychoacoustic model wasn't tuned on. The encoder isn't cheating; it's optimized for music, and some sounds barely are.

When you probably can't

On Bluetooth earbuds, in the car, on laptop speakers, in a noisy gym — the environment and the equipment are the bottleneck, not the codec. And here's the uncomfortable part: your brain desperately wants to hear a difference when it knows which file is which. Expectation bias is incredibly powerful. People reliably "hear" improvements when told they're listening to the expensive version of something — even when it's the exact same file. Which is exactly why blind testing matters.

Bluetooth deserves its own sentence, because it's a double layer of lossy: your music may arrive at the phone as a perfectly good file and still be re-encoded on its way to your ears, since Bluetooth audio itself runs through codecs like SBC, AAC, aptX, or LDAC. If you're judging "Spotify versus lossless" over wireless earbuds, you've been comparing two lossy chains the whole time. That's not an argument against Bluetooth — convenience is a real feature — just a reminder that the last hop often matters more than the first.

How to test yourself properly

If you want your own answer rather than the literature's, you can run a valid test at home with a little discipline:

• Start from the same lossless master. Encode your lossy copy yourself from a CD rip or purchase. Comparing files from different masters proves nothing — masters differ more than codecs do.
• Level-match. Even 0.3 dB skews verdicts toward the louder file. Use a tool that normalizes, or accept that your result includes loudness bias.
• Blind it properly. Foobar2000's ABX component is the classic rig; web-based blind testers work too. If you can see which file is playing, the test is measuring your expectations.
• Use enough trials. Guessing right 6 times out of 10 is a coin flip. Guessing right 90% of the time obviously means something in the statistical sense.
• Pick hard material. If you can tell on killer samples, try again on ordinary music. If you can only tell on castanets, your practical answer is "almost never." That's a fine and honest result.

One warning: training helps. Listeners who practice with codec artifacts — learning what pre-echo and spectral smearing sound like — get measurably better at spotting them, sometimes to their own regret. There are people who can't unhear 128 kbps anymore. Choose your hobbies carefully.

So why bother with lossless at all?

Fair question. If 320 kbps is transparent for most people most of the time, why care? Two reasons. First, archiving: lossless is a perfect master you can convert to anything, forever, with zero generation loss. Second, certainty: you never have to wonder whether that specific track, on that specific day, through that specific gear, is the one where you'd notice. The data is all there.

There's a third reason that's less mystical than it sounds: future processing. Every DSP step you might apply later — equalization, room correction, volume normalization, a remix, slowing a track down to learn a guitar part — works better on complete data, and some work dramatically better. Lossy artifacts that were inaudible in casual playback can be amplified into audibility by processing you haven't thought of yet. Lossless is the only format that's finished making decisions about what you're allowed to hear.

And of course, none of this matters if your "lossless" file was secretly upscaled from an MP3 — which happens more often than you'd think. That's the whole reason this site exists: run your files through the analyzer and the spectrum will show you the truth, no ears required.

The verdict

Can you tell the difference? Try a blind test and find out — it's genuinely fun. Just don't be surprised if the results humble you. They humble everyone.

Related reading: what "truly lossless" actually means, and the practical benefits of lossless beyond audibility.