What Is Lossless Data Compression?
By Lucian — builder & engineer, LK Forge
Lossless compression makes a file smaller and then gives every original byte back — decompress it and you have an exact copy, not an approximation. That is the whole promise, and it is testable: compress, decompress, and check the result matches the source bit-for-bit. We ran that test on a standard set of files with five real compressors, and measured how small each one goes. Below is what lossless actually means, a live demo you can run on your own text, and the numbers.
Try it: lossless in your browser
Type or paste anything below. Your browser compresses it with gzip (the DEFLATE algorithm, via the built-in CompressionStream API), then immediately decompresses it and checks the result against what you typed. Nothing leaves your device. Watch the size shrink on repetitive text — and notice the output always comes back identical.
How small can lossless go?
There is a naive limit called the order-0 entropy — the best you can do if you only look at how often each byte appears, ignoring order. For this corpus that floor is 45% of the original. Real compressors blow straight through it by modeling context and repetition: gzip reaches 26%, and modern codecs (xz, brotli) reach about 17.5%. Bigger bar = smaller file.
Whole-corpus ratio (original size ÷ compressed size); the % is the compressed size as a share of the original. The order-0 entropy floor is the per-byte limit; every real codec beats it by exploiting patterns across bytes.
Compressibility depends on the data
The same codec behaves completely differently on different files. The standout is kennedy.xls: a spreadsheet full of structure and repeated zeros, it drops to just 5% of its size with xz. A fax image (ptt5) compresses almost as hard. English prose sits around 30%, and the already-dense SPARC binary (sum) resists the most. Each row is one file; every number below survived a byte-for-byte round-trip check.
| File | Type | Size | gzip | bzip2 | xz | zstd | brotli |
|---|---|---|---|---|---|---|---|
| alice29.txt | English text | 149 KB | 36% | 28% | 32% | 32% | 31% |
| asyoulik.txt | English text | 122 KB | 39% | 32% | 36% | 36% | 34% |
| cp.html | HTML | 24 KB | 32% | 31% | 31% | 31% | 28% |
| fields.c | C source | 11 KB | 28% | 27% | 27% | 27% | 24% |
| grammar.lsp | Lisp source | 4 KB | 33% | 34% | 35% | 33% | 30% |
| kennedy.xls | Spreadsheet | 1006 KB | 20% | 13% | 5% | 6% | 6% |
| lcet10.txt | Technical text | 417 KB | 34% | 25% | 28% | 28% | 27% |
| plrabn12.txt | Poetry | 471 KB | 40% | 30% | 34% | 35% | 34% |
| ptt5 | Fax image | 501 KB | 10% | 10% | 8% | 9% | 8% |
| sum | SPARC binary | 37 KB | 33% | 34% | 25% | 29% | 27% |
| xargs.1 | Man page | 4 KB | 41% | 42% | 43% | 41% | 35% |
Each cell is the compressed size as a percent of the original (smaller is better). Already-compressed data — a JPEG, an MP3, a ZIP — has little redundancy left, so lossless tools barely shrink it; that is expected, not a failure.
When to use lossless vs lossy
Use lossless whenever every byte matters: text, source code, spreadsheets, logs, executables, and archives — anything where a changed bit is a bug. Use lossy when the data is headed for human eyes or ears and a small, well-chosen loss is invisible: photos, music, streaming video. The reason lossy wins so much more size is exactly that it is allowed to discard information — see how that trade plays out for photos in image compression: quality vs file size, and for documents in how PDF compression works.
Want to try it on real files? Our Image Compressor and PDF Compressor run entirely in your browser — nothing is uploaded.
Reproduce this
No cherry-picking — every figure is the whole-corpus total on a standard public file set, using each tool's stock maximum setting. Run it yourself:
- Corpus: the Canterbury corpus — 11 files, 2,810,784 bytes — from
corpus.canterbury.ac.nz/resources/cantrbry.zip(SHA-256c44b686d…39a1a6). - Settings:
gzip -9,bzip2 -9,xz -9e,zstd -19,brotli -q11— each at maximum. - Ratio: total original bytes ÷ total compressed bytes, summed over all 11 files.
- Lossless check: each stream was decompressed and compared to the original by SHA-256 — all matched.
# one file, every codec, at max — sizes in bytes
for f in corpus/*; do
o=$(wc -c < "$f")
g=$(gzip -9 -c "$f" | wc -c)
b=$(bzip2 -9 -c "$f" | wc -c)
x=$(xz -9e -c "$f" | wc -c)
z=$(zstd -19 -c "$f" | wc -c)
r=$(brotli -q 11 -c "$f" | wc -c)
echo "$f orig=$o gz=$g bz=$b xz=$x zstd=$z br=$r"
doneExact bytes depend on tool versions (here: gzip 1.14, bzip2 1.0.8, xz 5.8.3, zstd 1.5.7, brotli 1.2.0) — the ordering and the ~5.7× ceiling generalize. The order-0 entropy floor is -∑ p·log₂p over byte frequencies.
Common questions
Does lossless compression lose any quality?
No. Lossless compression recovers the original data bit-for-bit — decompressing a losslessly compressed file returns exactly the bytes you started with. In our test, all 55 file-and-codec round-trips matched the original by SHA-256. Quality loss only happens with lossy compression such as JPEG or MP3, which discard detail permanently to save more space.
Is ZIP lossless?
Yes. ZIP (and gzip) use the DEFLATE algorithm, which is lossless — the extracted files are identical to the originals. So are PNG, FLAC, GIF, and general-purpose tools like 7-Zip, xz, zstd and brotli. JPEG, MP3, AAC and H.264 are lossy.
What is the difference between lossless and lossy compression?
Lossless compression is perfectly reversible: you get every original byte back. Lossy compression throws away information the algorithm judges unimportant (fine image detail, inaudible sound) and cannot reconstruct it, which is why it reaches much smaller sizes. Use lossless for text, code, spreadsheets and archives; lossy for photos, music and video where a little loss is acceptable.
Which lossless compressor is best?
For the smallest file, brotli -q11 and xz -9e were the strongest in our test (about 5.7x on the Canterbury corpus, versus gzip's 3.85x). zstd -19 was nearly as small and much faster to compress. gzip remains the most universally supported. There is no single winner — it is a trade of size against speed and compatibility.
How much can lossless compression shrink a file?
It depends entirely on the data's redundancy. Highly structured files compress hugely — a spreadsheet in our test dropped to 5% of its size with xz — while already-random or already-compressed data barely shrinks at all. Across a mixed corpus, modern codecs reached about 17-18% of the original size.