diff --git a/src/transformers/tokenization_utils_base.py b/src/transformers/tokenization_utils_base.py
index 267d72485f..cf12f2d720 100644
--- a/src/transformers/tokenization_utils_base.py
+++ b/src/transformers/tokenization_utils_base.py
@@ -1562,6 +1562,8 @@ class PreTrainedTokenizerBase(SpecialTokensMixin):
             for key, value in special_tokens_map.items():
                 if isinstance(value, dict):
                     value = AddedToken(**value)
+                elif isinstance(value, list):
+                    value = [AddedToken(**token) if isinstance(token, dict) else token for token in value]
                 setattr(tokenizer, key, value)
 
         # Add supplementary tokens.
@@ -1633,6 +1635,10 @@ class PreTrainedTokenizerBase(SpecialTokensMixin):
             for key, value in self.special_tokens_map_extended.items():
                 if isinstance(value, AddedToken):
                     write_dict[key] = value.__getstate__()
+                elif isinstance(value, list):
+                    write_dict[key] = [
+                        token.__getstate__() if isinstance(token, AddedToken) else token for token in value
+                    ]
                 else:
                     write_dict[key] = value
             f.write(json.dumps(write_dict, ensure_ascii=False))
diff --git a/tests/test_tokenization_common.py b/tests/test_tokenization_common.py
index ba891f0cbb..4b841f850e 100644
--- a/tests/test_tokenization_common.py
+++ b/tests/test_tokenization_common.py
@@ -1165,6 +1165,16 @@ class TokenizerTesterMixin:
                         encoded_sequences_batch_padded_1[key], encoded_sequences_batch_padded_2[key],
                     )
 
+    def test_added_token_serializable(self):
+        tokenizers = self.get_tokenizers(do_lower_case=False)
+        for tokenizer in tokenizers:
+            new_token = AddedToken("new_token", lstrip=True)
+            tokenizer.add_special_tokens({"additional_special_tokens": [new_token]})
+
+            with tempfile.TemporaryDirectory() as tmp_dir_name:
+                tokenizer.save_pretrained(tmp_dir_name)
+                tokenizer.from_pretrained(tmp_dir_name)
+
     def test_batch_encode_plus_padding(self):
         # Test that padded sequences are equivalent between batch_encode_plus and encode_plus