diff --git a/doc/source/whatsnew/v1.4.0.rst b/doc/source/whatsnew/v1.4.0.rst
index be91020d357b6..8608dffd58090 100644
--- a/doc/source/whatsnew/v1.4.0.rst
+++ b/doc/source/whatsnew/v1.4.0.rst
@@ -216,8 +216,9 @@ Interval
 
 Indexing
 ^^^^^^^^
-- Bug in indexing on a :class:`Series` or :class:`DataFrame` with a :class:`DatetimeIndex` when passing a string, the return type depended on whether the index was monotonic (:issue:`24892`)
 - Bug in :meth:`DataFrame.truncate` and :meth:`Series.truncate` when the object's Index has a length greater than one but only one unique value (:issue:`42365`)
+- Bug in :meth:`Series.loc` when with a :class:`MultiIndex` whose first level contains only ``np.nan`` values (:issue:`42055`)
+- Bug in indexing on a :class:`Series` or :class:`DataFrame` with a :class:`DatetimeIndex` when passing a string, the return type depended on whether the index was monotonic (:issue:`24892`)
 - Bug in indexing on a :class:`MultiIndex` failing to drop scalar levels when the indexer is a tuple containing a datetime-like string (:issue:`42476`)
 -
 
diff --git a/pandas/core/indexes/base.py b/pandas/core/indexes/base.py
index 3b30af4bb6b1e..244bc2465f25c 100644
--- a/pandas/core/indexes/base.py
+++ b/pandas/core/indexes/base.py
@@ -1886,14 +1886,21 @@ def _drop_level_numbers(self, levnums: list[int]):
             new_names.pop(i)
 
         if len(new_levels) == 1:
+            lev = new_levels[0]
 
-            # set nan if needed
-            mask = new_codes[0] == -1
-            result = new_levels[0].take(new_codes[0])
-            if mask.any():
-                result = result.putmask(mask, np.nan)
+            if len(lev) == 0:
+                # If lev is empty, lev.take will fail GH#42055
+                res_values = algos.take(lev._values, new_codes[0], allow_fill=True)
+                result = type(lev)._simple_new(res_values, name=new_names[0])
+            else:
+                # set nan if needed
+                mask = new_codes[0] == -1
+                result = new_levels[0].take(new_codes[0])
+                if mask.any():
+                    result = result.putmask(mask, np.nan)
+
+                result._name = new_names[0]
 
-            result._name = new_names[0]
             return result
         else:
             from pandas.core.indexes.multi import MultiIndex
diff --git a/pandas/tests/indexing/test_loc.py b/pandas/tests/indexing/test_loc.py
index 2e0f48849530a..6de83e34122c2 100644
--- a/pandas/tests/indexing/test_loc.py
+++ b/pandas/tests/indexing/test_loc.py
@@ -1663,6 +1663,18 @@ def test_loc_multiindex_levels_contain_values_not_in_index_anymore(self, lt_valu
         with pytest.raises(KeyError, match=r"\['b'\] not in index"):
             df.loc[df["a"] < lt_value, :].loc[["b"], :]
 
+    def test_loc_multiindex_null_slice_na_level(self):
+        # GH#42055
+        lev1 = np.array([np.nan, np.nan])
+        lev2 = ["bar", "baz"]
+        mi = MultiIndex.from_arrays([lev1, lev2])
+        ser = Series([0, 1], index=mi)
+        result = ser.loc[:, "bar"]
+
+        # TODO: should we have name="bar"?
+        expected = Series([0], index=[np.nan])
+        tm.assert_series_equal(result, expected)
+
     def test_loc_drops_level(self):
         # Based on test_series_varied_multiindex_alignment, where
         #  this used to fail to drop the first level