diff --git a/doc/source/whatsnew/v1.6.0.rst b/doc/source/whatsnew/v1.6.0.rst
index 405b8cc0a5ded..ea3ba77dcca0f 100644
--- a/doc/source/whatsnew/v1.6.0.rst
+++ b/doc/source/whatsnew/v1.6.0.rst
@@ -32,6 +32,7 @@ Other enhancements
 - :meth:`Series.add_suffix`, :meth:`DataFrame.add_suffix`, :meth:`Series.add_prefix` and :meth:`DataFrame.add_prefix` support an ``axis`` argument. If ``axis`` is set, the default behaviour of which axis to consider can be overwritten (:issue:`47819`)
 - :func:`assert_frame_equal` now shows the first element where the DataFrames differ, analogously to ``pytest``'s output (:issue:`47910`)
 - Added ``index`` parameter to :meth:`DataFrame.to_dict` (:issue:`46398`)
+- Added metadata propagation for binary operators on :class:`DataFrame` (:issue:`28283`)
 -
 
 .. ---------------------------------------------------------------------------
diff --git a/pandas/core/frame.py b/pandas/core/frame.py
index 22ccd1d763769..447feb6986588 100644
--- a/pandas/core/frame.py
+++ b/pandas/core/frame.py
@@ -7711,7 +7711,7 @@ def _construct_result(self, result) -> DataFrame:
         -------
         DataFrame
         """
-        out = self._constructor(result, copy=False)
+        out = self._constructor(result, copy=False).__finalize__(self)
         # Pin columns instead of passing to constructor for compat with
         #  non-unique columns case
         out.columns = self.columns
diff --git a/pandas/tests/generic/test_duplicate_labels.py b/pandas/tests/generic/test_duplicate_labels.py
index 0546534d91399..c9036958cbd74 100644
--- a/pandas/tests/generic/test_duplicate_labels.py
+++ b/pandas/tests/generic/test_duplicate_labels.py
@@ -68,9 +68,7 @@ def test_to_frame(self):
         assert ser.to_frame().flags.allows_duplicate_labels is False
 
     @pytest.mark.parametrize("func", ["add", "sub"])
-    @pytest.mark.parametrize(
-        "frame", [False, pytest.param(True, marks=not_implemented)]
-    )
+    @pytest.mark.parametrize("frame", [False, True])
     @pytest.mark.parametrize("other", [1, pd.Series([1, 2], name="A")])
     def test_binops(self, func, other, frame):
         df = pd.Series([1, 2], name="A", index=["a", "b"]).set_flags(
diff --git a/pandas/tests/generic/test_finalize.py b/pandas/tests/generic/test_finalize.py
index dddab05af7341..af1a76a6263c7 100644
--- a/pandas/tests/generic/test_finalize.py
+++ b/pandas/tests/generic/test_finalize.py
@@ -109,13 +109,10 @@
     (pd.DataFrame, frame_data, operator.methodcaller("nlargest", 1, "A")),
     (pd.DataFrame, frame_data, operator.methodcaller("nsmallest", 1, "A")),
     (pd.DataFrame, frame_mi_data, operator.methodcaller("swaplevel")),
-    pytest.param(
-        (
-            pd.DataFrame,
-            frame_data,
-            operator.methodcaller("add", pd.DataFrame(*frame_data)),
-        ),
-        marks=not_implemented_mark,
+    (
+        pd.DataFrame,
+        frame_data,
+        operator.methodcaller("add", pd.DataFrame(*frame_data)),
     ),
     # TODO: div, mul, etc.
     pytest.param(
@@ -539,21 +536,18 @@ def test_finalize_called_eval_numexpr():
         (pd.DataFrame({"A": [1]}), pd.Series([1])),
     ],
 )
-def test_binops(request, args, annotate, all_arithmetic_functions):
-    # This generates 326 tests... Is that needed?
+def test_binops(request, args, annotate, all_binary_operators):
+    # This generates 624 tests... Is that needed?
     left, right = args
     if annotate == "both" and isinstance(left, int) or isinstance(right, int):
         return
 
-    if isinstance(left, pd.DataFrame) or isinstance(right, pd.DataFrame):
-        request.node.add_marker(pytest.mark.xfail(reason="not implemented"))
-
     if annotate in {"left", "both"} and not isinstance(left, int):
         left.attrs = {"a": 1}
     if annotate in {"left", "both"} and not isinstance(right, int):
         right.attrs = {"a": 1}
 
-    result = all_arithmetic_functions(left, right)
+    result = all_binary_operators(left, right)
     assert result.attrs == {"a": 1}