diff --git a/docs/linter.md b/docs/linter.md
index 98546538a5..1f3e4f22b5 100644
--- a/docs/linter.md
+++ b/docs/linter.md
@@ -337,6 +337,27 @@ violations on a single line.
 Note that Ruff will also respect Flake8's `# flake8: noqa` directive, and will treat it as
 equivalent to `# ruff: noqa`.
 
+### Full suppression comment specification
+
+The full specification is as follows:
+
+- An inline blanket `noqa` comment is given by a case-insensitive match for
+  `#noqa` with optional whitespace after the `#` symbol, followed by either: the
+  end of the comment, the beginning of a new comment (`#`), or whitespace
+  followed by any character other than `:`.
+- An inline rule suppression is given by first finding a case-insensitive match
+  for `#noqa` with optional whitespace after the `#` symbol, optional whitespace
+  after `noqa`, and followed by the symbol `:`. After this we are expected to
+  have a list of rule codes which is given by sequences of uppercase ASCII
+  characters followed by ASCII digits, separated by whitespace or commas. The
+  list ends at the last valid code. We will attempt to interpret rules with a
+  missing delimiter (e.g. `F401F841`), though a warning will be emitted in this
+  case.
+- A file-level exemption comment is given by a case-sensitive match for `#ruff:`
+  or `#flake8:`, with optional whitespace after `#` and before `:`, followed by
+  optional whitespace and a case-insensitive match for `noqa`. After this, the
+  specification is as in the inline case.
+
 ### Detecting unused suppression comments
 
 Ruff implements a special rule, [`unused-noqa`](https://docs.astral.sh/ruff/rules/unused-noqa/),