diff --git a/doc/changes/devel.rst b/doc/changes/devel.rst
index 729915a453b..a639a64428b 100644
--- a/doc/changes/devel.rst
+++ b/doc/changes/devel.rst
@@ -56,7 +56,7 @@ Bugs
 - Fix bug with axis clip box boundaries in :func:`mne.viz.plot_evoked_topo` and related functions (:gh:`11999` by `Eric Larson`_)
 - Fix bug with ``subject_info`` when loading data from and exporting to EDF file (:gh:`11952` by `Paul Roujansky`_)
 - Fix bug with delayed checking of :class:`info["bads"] <mne.Info>` (:gh:`12038` by `Eric Larson`_)
-- Fix handling of channel information in annotations when loading data from and exporting to EDF file (:gh:`11960` :gh:`12017` by `Paul Roujansky`_)
+- Fix handling of channel information in annotations when loading data from and exporting to EDF file (:gh:`11960` :gh:`12017` :gh:`12044` by `Paul Roujansky`_)
 - Add missing ``overwrite`` and ``verbose`` parameters to :meth:`Transform.save() <mne.transforms.Transform.save>` (:gh:`12004` by `Marijn van Vliet`_)
 - Fix parsing of eye-link :class:`~mne.Annotations` when ``apply_offsets=False`` is provided to :func:`~mne.io.read_raw_eyelink` (:gh:`12003` by `Mathieu Scheltienne`_)
 - Correctly prune channel-specific :class:`~mne.Annotations` when creating :class:`~mne.Epochs` without the channel(s) included in the channel specific annotations (:gh:`12010` by `Mathieu Scheltienne`_)
diff --git a/mne/io/edf/edf.py b/mne/io/edf/edf.py
index dd8a80d3fef..a1dce08c049 100644
--- a/mne/io/edf/edf.py
+++ b/mne/io/edf/edf.py
@@ -206,6 +206,7 @@ def __init__(
             )
             annotations = _read_annotations_edf(
                 tal_data[0],
+                ch_names=info["ch_names"],
                 encoding=encoding,
             )
             self.set_annotations(annotations, on_missing="warn")
@@ -1892,25 +1893,21 @@ def read_raw_gdf(
 
 
 @fill_doc
-def _read_annotations_edf(annotations, encoding="utf8"):
+def _read_annotations_edf(annotations, ch_names=None, encoding="utf8"):
     """Annotation File Reader.
 
     Parameters
     ----------
     annotations : ndarray (n_chans, n_samples) | str
         Channel data in EDF+ TAL format or path to annotation file.
+    ch_names : list of string
+        List of channels' names.
     %(encoding_edf)s
 
     Returns
     -------
-    onset : array of float, shape (n_annotations,)
-        The starting time of annotations in seconds after ``orig_time``.
-    duration : array of float, shape (n_annotations,)
-        Durations of the annotations in seconds.
-    description : array of str, shape (n_annotations,)
-        Array of strings containing description for each annotation. If a
-        string, all the annotations are given the same description. To reject
-        epochs, use description starting with keyword 'bad'. See example above.
+    annot : instance of Annotations
+        The annotations.
     """
     pat = "([+-]\\d+\\.?\\d*)(\x15(\\d+\\.?\\d*))?(\x14.*?)\x14\x00"
     if isinstance(annotations, str):
@@ -1949,7 +1946,11 @@ def _read_annotations_edf(annotations, encoding="utf8"):
         duration = float(ev[2]) if ev[2] else 0
         for description in ev[3].split("\x14")[1:]:
             if description:
-                if "@@" in description:
+                if (
+                    "@@" in description
+                    and ch_names is not None
+                    and description.split("@@")[1] in ch_names
+                ):
                     description, ch_name = description.split("@@")
                     key = f"{onset}_{duration}_{description}"
                 else:
@@ -1979,22 +1980,20 @@ def _read_annotations_edf(annotations, encoding="utf8"):
                 offset = -onset
 
     if events:
-        onset, duration, description, ch_names = zip(*events.values())
+        onset, duration, description, annot_ch_names = zip(*events.values())
     else:
-        onset, duration, description, ch_names = list(), list(), list(), list()
+        onset, duration, description, annot_ch_names = list(), list(), list(), list()
 
-    assert len(onset) == len(duration) == len(description) == len(ch_names)
+    assert len(onset) == len(duration) == len(description) == len(annot_ch_names)
 
-    annotations = Annotations(
+    return Annotations(
         onset=onset,
         duration=duration,
         description=description,
         orig_time=None,
-        ch_names=ch_names,
+        ch_names=annot_ch_names,
     )
 
-    return annotations
-
 
 def _get_annotations_gdf(edf_info, sfreq):
     onset, duration, desc = list(), list(), list()
diff --git a/mne/io/edf/tests/test_edf.py b/mne/io/edf/tests/test_edf.py
index 9fad34c064b..e1c176c7e4c 100644
--- a/mne/io/edf/tests/test_edf.py
+++ b/mne/io/edf/tests/test_edf.py
@@ -24,7 +24,7 @@
 import pytest
 
 from mne import pick_types, Annotations
-from mne.annotations import events_from_annotations, read_annotations
+from mne.annotations import _ndarray_ch_names, events_from_annotations, read_annotations
 from mne.datasets import testing
 from mne.io import read_raw_edf, read_raw_bdf, read_raw_fif, edf, read_raw_gdf
 from mne.io.tests.test_raw import _test_raw_reader
@@ -504,6 +504,81 @@ def test_read_utf8_annotations():
     assert raw.annotations[1]["description"] == "仰卧"
 
 
+def test_read_annotations_edf(tmp_path):
+    """Test reading annotations from EDF file."""
+    annot = (
+        b"+1.1\x14Event A@@CH1\x14\x00\x00"
+        b"+1.2\x14Event A\x14\x00\x00"
+        b"+1.3\x14Event B@@CH1\x14\x00\x00"
+        b"+1.3\x14Event B@@CH2\x14\x00\x00"
+        b"+1.4\x14Event A@@CH3\x14\x00\x00"
+        b"+1.5\x14Event B\x14\x00\x00"
+    )
+    annot_file = tmp_path / "annotations.edf"
+    with open(annot_file, "wb") as f:
+        f.write(annot)
+
+    # Test reading annotations from channel data
+    with open(annot_file, "rb") as f:
+        tal_channel = _read_ch(
+            f,
+            subtype="EDF",
+            dtype="<i2",
+            samp=-1,
+            dtype_byte=None,
+        )
+
+    # Read annotations without input channel names: annotations are left untouched and
+    # assigned as global
+    annotations = _read_annotations_edf(tal_channel, ch_names=None, encoding="latin1")
+    assert_allclose(annotations.onset, [1.1, 1.2, 1.3, 1.3, 1.4, 1.5])
+    assert not any(annotations.duration)  # all durations are 0
+    assert_array_equal(
+        annotations.description,
+        [
+            "Event A@@CH1",
+            "Event A",
+            "Event B@@CH1",
+            "Event B@@CH2",
+            "Event A@@CH3",
+            "Event B",
+        ],
+    )
+    assert_array_equal(
+        annotations.ch_names, _ndarray_ch_names([(), (), (), (), (), ()])
+    )
+
+    # Read annotations with complete input channel names: each annotation is parsed and
+    # associated to a channel
+    annotations = _read_annotations_edf(
+        tal_channel, ch_names=["CH1", "CH2", "CH3"], encoding="latin1"
+    )
+    assert_allclose(annotations.onset, [1.1, 1.2, 1.3, 1.4, 1.5])
+    assert not any(annotations.duration)  # all durations are 0
+    assert_array_equal(
+        annotations.description, ["Event A", "Event A", "Event B", "Event A", "Event B"]
+    )
+    assert_array_equal(
+        annotations.ch_names,
+        _ndarray_ch_names([("CH1",), (), ("CH1", "CH2"), ("CH3",), ()]),
+    )
+
+    # Read annotations with incomplete input channel names: "CH3" is missing from input
+    # channels, turning the related annotation into a global one
+    annotations = _read_annotations_edf(
+        tal_channel, ch_names=["CH1", "CH2"], encoding="latin1"
+    )
+    assert_allclose(annotations.onset, [1.1, 1.2, 1.3, 1.4, 1.5])
+    assert not any(annotations.duration)  # all durations are 0
+    assert_array_equal(
+        annotations.description,
+        ["Event A", "Event A", "Event B", "Event A@@CH3", "Event B"],
+    )
+    assert_array_equal(
+        annotations.ch_names, _ndarray_ch_names([("CH1",), (), ("CH1", "CH2"), (), ()])
+    )
+
+
 def test_read_latin1_annotations(tmp_path):
     """Test if annotations encoded as Latin-1 can be read.