diff --git a/spacy/matcher/matcher.pyx b/spacy/matcher/matcher.pyx
index 644f7704b..a367dcc3a 100644
--- a/spacy/matcher/matcher.pyx
+++ b/spacy/matcher/matcher.pyx
@@ -313,7 +313,8 @@ cdef find_matches(TokenPatternC** patterns, int n, object doclike, int length, e
         # We need to deduplicate, because we could otherwise arrive at the same
         # match through two paths, e.g. .?.? matching 'a'. Are we matching the
         # first .?, or the second .? -- it doesn't matter, it's just one match.
-        if match not in seen:
+        # Skip 0-length matches. (TODO: fix algorithm)
+        if match not in seen and matches[i].length > 0:
             output.append(match)
             seen.add(match)
     return output
diff --git a/spacy/tests/matcher/test_matcher_api.py b/spacy/tests/matcher/test_matcher_api.py
index 236f25130..75ee255d4 100644
--- a/spacy/tests/matcher/test_matcher_api.py
+++ b/spacy/tests/matcher/test_matcher_api.py
@@ -493,3 +493,13 @@ def test_matcher_remove_zero_operator(en_vocab):
     assert "Rule" in matcher
     matcher.remove("Rule")
     assert "Rule" not in matcher
+
+
+def test_matcher_no_zero_length(en_vocab):
+    doc = Doc(en_vocab, words=["a", "b"])
+    doc[0].tag_ = "A"
+    doc[1].tag_ = "B"
+    doc.is_tagged = True
+    matcher = Matcher(en_vocab)
+    matcher.add("TEST", [[{"TAG": "C", "OP": "?"}]])
+    assert len(matcher(doc)) == 0