diff --git a/docs/source/plugin-development/index.rst b/docs/source/plugin-development/index.rst
index 9088942..50a88a0 100644
--- a/docs/source/plugin-development/index.rst
+++ b/docs/source/plugin-development/index.rst
@@ -43,6 +43,8 @@ section we cover:
 
 - How to write your check plugin so that it works with |Flake8| 2.x and 3.x
 
+- How to test your plugin (:doc:`testing-plugins`)
+
 
 Video Tutorial
 ==============
@@ -65,3 +67,4 @@ Detailed Plugin Development Documentation
     registering-plugins
     plugin-parameters
     formatters
+    testing-plugins
diff --git a/docs/source/plugin-development/testing-plugins.rst b/docs/source/plugin-development/testing-plugins.rst
new file mode 100644
index 0000000..459e154
--- /dev/null
+++ b/docs/source/plugin-development/testing-plugins.rst
@@ -0,0 +1,91 @@
+=================
+ Testing plugins
+=================
+
+This section describes practical ways to verify that a |Flake8| plugin behaves
+the way you expect. It complements :ref:`register-a-plugin`, which explains
+how to expose your code to |Flake8| via entry points.
+
+Install the package you are testing (for example with ``pip install -e .`` in
+your project root) so metadata and entry points are visible to |Flake8| in the
+same environment that runs your tests.
+
+.. note::
+
+   The programmatic API lives in :mod:`flake8.api.legacy`. For background and
+   full reference, see :doc:`../user/python-api`.
+
+   :meth:`~flake8.api.legacy.StyleGuide.input_file` **does not** accept in-memory
+   ``lines`` anymore: since |Flake8| 3.0 those arguments are ignored, and only
+   the path is checked. For tests, write source to a temporary file and pass
+   that path to :meth:`~flake8.api.legacy.StyleGuide.check_files`.
+
+Programmatic checks
+===================
+
+Use :func:`~flake8.api.legacy.get_style_guide` to build a
+:class:`~flake8.api.legacy.StyleGuide`, then call
+:meth:`~flake8.api.legacy.StyleGuide.check_files` with a list of paths. The
+returned :class:`~flake8.api.legacy.Report` exposes
+:attr:`~flake8.api.legacy.Report.total_errors` and
+:meth:`~flake8.api.legacy.Report.get_statistics`.
+
+Limit reported codes to those from your plugin (prefix matching works the same
+as for ``--select`` on the command line) so unrelated built-in violations do not
+make assertions brittle:
+
+.. code-block:: python
+
+    from pathlib import Path
+
+    from flake8.api import legacy
+
+
+    def test_plugin_reports_violation(tmp_path):
+        example = tmp_path / "example.py"
+        example.write_text("# plugin-specific bad pattern\n", encoding="utf-8")
+
+        style_guide = legacy.get_style_guide(select=["X"])
+        report = style_guide.check_files([str(example)])
+
+        assert report.total_errors == 1
+
+Run the full default |Flake8| selection when you want an integration-style test
+that your plugin plays nicely with everything else a user would run:
+
+.. code-block:: python
+
+    style_guide = legacy.get_style_guide()
+    report = style_guide.check_files([str(example)])
+
+Keyword arguments to :func:`~flake8.api.legacy.get_style_guide` correspond to
+parsed command-line options on the application (for example ``select``,
+``ignore``, ``exclude``). Only attributes that already exist on the options
+object can be set; unknown names are logged and ignored.
+
+Command-line checks
+===================
+
+End-to-end tests can invoke the same entry point users run:
+
+.. code-block:: python
+
+    import subprocess
+    import sys
+
+
+    def test_flake8_cli(tmp_path):
+        target = tmp_path / "example.py"
+        target.write_text("...\n", encoding="utf-8")
+
+        result = subprocess.run(
+            [sys.executable, "-m", "flake8", "--select=X", str(target)],
+            check=False,
+            capture_output=True,
+            text=True,
+        )
+        assert result.returncode == 1
+        assert "X" in result.stdout
+
+Use ``--select`` (or ``--extend-select``) to focus on your plugin's codes when
+asserting on output text.