diff --git a/doc/source/whatsnew/v0.18.1.txt b/doc/source/whatsnew/v0.18.1.txt
index f20b961455ba7..f6873cca02dbe 100644
--- a/doc/source/whatsnew/v0.18.1.txt
+++ b/doc/source/whatsnew/v0.18.1.txt
@@ -211,3 +211,4 @@ Bug Fixes
 - Bug in ``.describe()`` resets categorical columns information (:issue:`11558`)
 - Bug where ``loffset`` argument was not applied when calling ``resample().count()`` on a timeseries (:issue:`12725`)
 - ``pd.read_excel()`` now accepts path objects (e.g. ``pathlib.Path``, ``py.path.local``) for the file path, in line with other ``read_*`` functions (:issue:`12655`)
+- Bug in ``GroupBy.filter`` when ``dropna=False`` and no groups fulfilled the criteria (:issue:`12768`) 
diff --git a/pandas/core/groupby.py b/pandas/core/groupby.py
index 066afc55e442f..cde7c40054847 100644
--- a/pandas/core/groupby.py
+++ b/pandas/core/groupby.py
@@ -820,7 +820,7 @@ def _concat_objects(self, keys, values, not_indexed_same=False):
 
     def _apply_filter(self, indices, dropna):
         if len(indices) == 0:
-            indices = []
+            indices = np.array([])
         else:
             indices = np.sort(np.concatenate(indices))
         if dropna:
diff --git a/pandas/tests/test_groupby.py b/pandas/tests/test_groupby.py
index 28038e02b64ca..bfad309b6cd50 100644
--- a/pandas/tests/test_groupby.py
+++ b/pandas/tests/test_groupby.py
@@ -4990,6 +4990,21 @@ def test_filter_out_no_groups(self):
         filtered = grouped.filter(lambda x: x['A'].mean() > 0)
         assert_frame_equal(filtered, df)
 
+    def test_filter_out_all_groups_in_df_dropna_false(self):
+        # GH12768
+        df = pd.DataFrame({'a': [1, 1, 2], 'b': [1, 2, 0]})
+        res = df.groupby('a')
+        res = res.filter(lambda x: x['b'].sum() > 5, dropna=False)
+        expected = pd.DataFrame({'a': [nan] * 3, 'b': [nan] * 3})
+        assert_frame_equal(expected, res)
+
+    def test_filter_out_all_groups_in_df_dropna_true(self):
+        df = pd.DataFrame({'a': [1, 1, 2], 'b': [1, 2, 0]})
+        res = df.groupby('a')
+        res = res.filter(lambda x: x['b'].sum() > 5, dropna=True)
+        expected = pd.DataFrame({'a': [], 'b': []}, dtype="int64")
+        assert_frame_equal(expected, res)
+
     def test_filter_condition_raises(self):
         def raise_if_sum_is_zero(x):
             if x.sum() == 0: