.LZMA File Format – Header

LZMA compressed file

An .lzma file is a 13-byte header followed by one LZMA stream. This page follows lzma-specification.txt from Igor Pavlov’s LZMA SDK (a draft, read 2026-11-12 from a mirror of the SDK’s DOC folder). Numbers are little-endian. The compressed data is described on LZMA coding.

Fields

OffsetSizeNameDescription
01Properties(pb × 5 + lp) × 9 + lc; must be less than 225. See properties.
14Dictionary sizeIn bytes. A decoder uses at least 4,096 (212) even if a smaller value is written.
58Uncompressed sizeSize of the decompressed data, or FF FF FF FF FF FF FF FF when unknown.
13…LZMA dataThe range-coded stream.

Properties

To unpack the byte d: lc = d % 9, d = d / 9, lp = d % 5, pb = d / 5.

NameRangeMeaning
lc0 to 8Literal context bits: how many high bits of the previous byte choose the model for the next literal.
lp0 to 4Literal position bits: how many low bits of the position choose it too.
pb0 to 4Position bits: how many low bits of the position choose the models for match and repeat decisions.

The byte must be below 225 (9 × 5 × 5); a decoder rejects anything else. lc 3, lp 0, pb 2 give 5D; the specification calls lc 3 and lp 0 the default parameters. The format has no signature.

Where the data ends

  • With a known size, the decoder stops after that many bytes. The stream may still end with an end marker, which is then optional.
  • With an unknown size, the stream must end with the end marker: a match whose distance decodes to FFFFFFFF.
  • After the last bit the range decoder’s Code must be 0; the specification calls this check a very good way to detect corruption.

The format has no checksum, so damage may go unnoticed. The same five property bytes (properties and dictionary size) are used in 7z for the LZMA coder, without the size.

Example

5D                         properties: lc 3, lp 0, pb 2
00 00 80 00                dictionary 8 MiB (0x00800000)
FF FF FF FF FF FF FF FF    uncompressed size unknown: an end marker follows the data
00 …                       LZMA data; the first byte of the range coder is always 00