diff --git a/rest_framework/filters.py b/rest_framework/filters.py
index 213182dad..366577519 100644
--- a/rest_framework/filters.py
+++ b/rest_framework/filters.py
@@ -85,7 +85,7 @@ class SearchFilter(BaseFilterBackend):
                 search_field = search_field[1:]
             # Annotated fields do not need to be distinct
             if isinstance(queryset, models.QuerySet) and search_field in queryset.query.annotations:
-                return False
+                continue
             parts = search_field.split(LOOKUP_SEP)
             for part in parts:
                 field = opts.get_field(part)
diff --git a/tests/test_filters.py b/tests/test_filters.py
index 30cedc7d7..e69537666 100644
--- a/tests/test_filters.py
+++ b/tests/test_filters.py
@@ -406,6 +406,21 @@ class SearchFilterAnnotatedFieldTests(TestCase):
         assert len(response.data) == 1
         assert response.data[0]['title_text'] == 'ABCDEF'
 
+    def test_must_call_distinct_subsequent_m2m_fields(self):
+        f = filters.SearchFilter()
+
+        queryset = SearchFilterModelM2M.objects.annotate(
+            title_text=Upper(
+                Concat(models.F('title'), models.F('text'))
+            )
+        ).all()
+
+        # Sanity check that m2m must call distinct
+        assert f.must_call_distinct(queryset, ['attributes'])
+
+        # Annotated field should not prevent m2m must call distinct
+        assert f.must_call_distinct(queryset, ['title_text', 'attributes'])
+
 
 class OrderingFilterModel(models.Model):
     title = models.CharField(max_length=20, verbose_name='verbose title')