diff --git a/_pytest/mark/legacy.py b/_pytest/mark/legacy.py
index ec45f12afdb..5c7b8d0013a 100644
--- a/_pytest/mark/legacy.py
+++ b/_pytest/mark/legacy.py
@@ -5,8 +5,6 @@
 import attr
 import keyword
 
-from . import MarkInfo, MarkDecorator
-
 from _pytest.config import UsageError
 
 
@@ -18,11 +16,8 @@ class MarkMapping(object):
     own_mark_names = attr.ib()
 
     @classmethod
-    def from_keywords(cls, keywords):
-        mark_names = set()
-        for key, value in keywords.items():
-            if isinstance(value, MarkInfo) or isinstance(value, MarkDecorator):
-                mark_names.add(key)
+    def from_item(cls, item):
+        mark_names = set(mark.name for mark in item.iter_markers())
         return cls(mark_names)
 
     def __getitem__(self, name):
@@ -70,7 +65,7 @@ def __getitem__(self, subname):
 
 def matchmark(colitem, markexpr):
     """Tries to match on any marker names, attached to the given colitem."""
-    return eval(markexpr, {}, MarkMapping.from_keywords(colitem.keywords))
+    return eval(markexpr, {}, MarkMapping.from_item(colitem))
 
 
 def matchkeyword(colitem, keywordexpr):
diff --git a/changelog/3441.bugfix b/changelog/3441.bugfix
new file mode 100644
index 00000000000..6afeeab6245
--- /dev/null
+++ b/changelog/3441.bugfix
@@ -0,0 +1 @@
+Also use iter_marker for discovering the marks applying for marker expressions from the cli to avoid the bad data from the legacy mark storage.
\ No newline at end of file
diff --git a/testing/test_mark.py b/testing/test_mark.py
index 9ec1ce75a99..31d3af3e52c 100644
--- a/testing/test_mark.py
+++ b/testing/test_mark.py
@@ -295,7 +295,7 @@ def test_mark_option_custom(spec, testdir):
         def pytest_collection_modifyitems(items):
             for item in items:
                 if "interface" in item.nodeid:
-                    item.keywords["interface"] = pytest.mark.interface
+                    item.add_marker(pytest.mark.interface)
     """)
     testdir.makepyfile("""
         def test_interface():
@@ -927,3 +927,35 @@ def test_parameterset_for_parametrize_marks(testdir, mark):
 def test_parameterset_for_parametrize_bad_markname(testdir):
     with pytest.raises(pytest.UsageError):
         test_parameterset_for_parametrize_marks(testdir, 'bad')
+
+
+def test_mark_expressions_no_smear(testdir):
+    testdir.makepyfile("""
+        import pytest
+
+        class BaseTests(object):
+            def test_something(self):
+                pass
+
+        @pytest.mark.FOO
+        class TestFooClass(BaseTests):
+            pass
+
+        @pytest.mark.BAR
+        class TestBarClass(BaseTests):
+            pass
+    """)
+
+    reprec = testdir.inline_run("-m", 'FOO')
+    passed, skipped, failed = reprec.countoutcomes()
+    dlist = reprec.getcalls("pytest_deselected")
+    assert passed == 1
+    assert skipped == failed == 0
+    deselected_tests = dlist[0].items
+    assert len(deselected_tests) == 1
+
+    # keywords smear - expected behaviour
+    reprec_keywords = testdir.inline_run("-k", 'FOO')
+    passed_k, skipped_k, failed_k = reprec_keywords.countoutcomes()
+    assert passed_k == 2
+    assert skipped_k == failed_k == 0