diff --git a/doc/source/whatsnew/v0.24.0.txt b/doc/source/whatsnew/v0.24.0.txt
index 90fc579ae69e5..a63276efc5b7c 100644
--- a/doc/source/whatsnew/v0.24.0.txt
+++ b/doc/source/whatsnew/v0.24.0.txt
@@ -17,7 +17,7 @@ Other Enhancements
 - :func:`to_datetime` now supports the ``%Z`` and ``%z`` directive when passed into ``format`` (:issue:`13486`)
 - :func:`Series.mode` and :func:`DataFrame.mode` now support the ``dropna`` parameter which can be used to specify whether NaN/NaT values should be considered (:issue:`17534`)
 - :func:`to_csv` now supports ``compression`` keyword when a file handle is passed. (:issue:`21227`)
-- :meth:`Index.droplevel` is now implemented also for flat indexes, for compatibility with MultiIndex (:issue:`21115`)
+- :meth:`Index.droplevel` is now implemented also for flat indexes, for compatibility with :class:`MultiIndex` (:issue:`21115`)
 
 
 .. _whatsnew_0240.api_breaking:
@@ -199,6 +199,7 @@ Indexing
 ^^^^^^^^
 
 - The traceback from a ``KeyError`` when asking ``.loc`` for a single missing label is now shorter and more clear (:issue:`21557`)
+- When ``.ix`` is asked for a missing integer label in a :class:`MultiIndex` with a first level of integer type, it now raises a ``KeyError`` - consistently with the case of a flat :class:`Int64Index` - rather than falling back to positional indexing (:issue:`21593`)
 -
 -
 
diff --git a/pandas/core/indexing.py b/pandas/core/indexing.py
index 1f9fe5f947d0c..a69313a2d4a43 100755
--- a/pandas/core/indexing.py
+++ b/pandas/core/indexing.py
@@ -13,7 +13,6 @@
     is_iterator,
     is_scalar,
     is_sparse,
-    _is_unorderable_exception,
     _ensure_platform_int)
 from pandas.core.dtypes.missing import isna, _infer_fill_value
 from pandas.errors import AbstractMethodError
@@ -139,10 +138,7 @@ def _get_label(self, label, axis=None):
             # as its basically direct indexing
             # but will fail when the index is not present
             # see GH5667
-            try:
-                return self.obj._xs(label, axis=axis)
-            except:
-                return self.obj[label]
+            return self.obj._xs(label, axis=axis)
         elif isinstance(label, tuple) and isinstance(label[axis], slice):
             raise IndexingError('no slices here, handle elsewhere')
 
@@ -1797,9 +1793,8 @@ class _LocIndexer(_LocationIndexer):
 
     @Appender(_NDFrameIndexer._validate_key.__doc__)
     def _validate_key(self, key, axis):
-        ax = self.obj._get_axis(axis)
 
-        # valid for a label where all labels are in the index
+        # valid for a collection of labels (we check their presence later)
         # slice of labels (where start-end in labels)
         # slice of integers (only if in the labels)
         # boolean
@@ -1807,32 +1802,11 @@ def _validate_key(self, key, axis):
         if isinstance(key, slice):
             return
 
-        elif com.is_bool_indexer(key):
+        if com.is_bool_indexer(key):
             return
 
-        elif not is_list_like_indexer(key):
-
-            def error():
-                if isna(key):
-                    raise TypeError("cannot use label indexing with a null "
-                                    "key")
-                raise KeyError(u"the label [{key}] is not in the [{axis}]"
-                               .format(key=key,
-                                       axis=self.obj._get_axis_name(axis)))
-
-            try:
-                key = self._convert_scalar_indexer(key, axis)
-            except TypeError as e:
-
-                # python 3 type errors should be raised
-                if _is_unorderable_exception(e):
-                    error()
-                raise
-            except:
-                error()
-
-            if not ax.contains(key):
-                error()
+        if not is_list_like_indexer(key):
+            self._convert_scalar_indexer(key, axis)
 
     def _is_scalar_access(self, key):
         # this is a shortcut accessor to both .loc and .iloc
diff --git a/pandas/tests/indexes/datetimes/test_partial_slicing.py b/pandas/tests/indexes/datetimes/test_partial_slicing.py
index 4580d9fff31d5..e1e80e50e31f0 100644
--- a/pandas/tests/indexes/datetimes/test_partial_slicing.py
+++ b/pandas/tests/indexes/datetimes/test_partial_slicing.py
@@ -11,6 +11,8 @@
                     date_range, Index, Timedelta, Timestamp)
 from pandas.util import testing as tm
 
+from pandas.core.indexing import IndexingError
+
 
 class TestSlicing(object):
     def test_dti_slicing(self):
@@ -313,12 +315,12 @@ def test_partial_slicing_with_multiindex(self):
         result = df_multi.loc[('2013-06-19 09:30:00', 'ACCT1', 'ABC')]
         tm.assert_series_equal(result, expected)
 
-        # this is a KeyError as we don't do partial string selection on
-        # multi-levels
+        # this is an IndexingError as we don't do partial string selection on
+        # multi-levels.
         def f():
             df_multi.loc[('2013-06-19', 'ACCT1', 'ABC')]
 
-        pytest.raises(KeyError, f)
+        pytest.raises(IndexingError, f)
 
         # GH 4294
         # partial slice on a series mi
diff --git a/pandas/tests/indexing/test_multiindex.py b/pandas/tests/indexing/test_multiindex.py
index 43656a392e582..d2c4c8f5e149b 100644
--- a/pandas/tests/indexing/test_multiindex.py
+++ b/pandas/tests/indexing/test_multiindex.py
@@ -230,7 +230,8 @@ def test_iloc_getitem_multiindex(self):
         # corner column
         rs = mi_int.iloc[2, 2]
         with catch_warnings(record=True):
-            xp = mi_int.ix[:, 2].ix[2]
+            # First level is int - so use .loc rather than .ix (GH 21593)
+            xp = mi_int.loc[(8, 12), (4, 10)]
         assert rs == xp
 
         # this is basically regular indexing
@@ -278,6 +279,12 @@ def test_loc_multiindex(self):
             xp = mi_int.ix[4]
         tm.assert_frame_equal(rs, xp)
 
+        # missing label
+        pytest.raises(KeyError, lambda: mi_int.loc[2])
+        with catch_warnings(record=True):
+            # GH 21593
+            pytest.raises(KeyError, lambda: mi_int.ix[2])
+
     def test_getitem_partial_int(self):
         # GH 12416
         # with single item