diff --git a/tests/fixtures/migration/cassette_with_duplicate_requests.yaml b/tests/fixtures/migration/cassette_with_duplicate_requests.yaml
new file mode 100644
index 00000000..121289ab
--- /dev/null
+++ b/tests/fixtures/migration/cassette_with_duplicate_requests.yaml
@@ -0,0 +1,19 @@
+interactions:
+- request:
+    body: null
+    headers:
+      accept: ['*/*']
+      accept-encoding: ['gzip, deflate, compress']
+      user-agent: ['python-requests/2.2.1 CPython/2.6.1 Darwin/10.8.0']
+    method: GET
+    uri: http://httpbin.org/ip
+  response:
+    body: {string: "{\n  \"origin\": \"217.122.164.194\"\n}"}
+    headers:
+      access-control-allow-origin: ['*']
+      content-type: [application/json]
+      date: ['Mon, 21 Apr 2014 23:06:09 GMT']
+      server: [gunicorn/0.17.4]
+      content-length: ['32']
+      connection: [keep-alive]
+    status: {code: 200, message: OK}
diff --git a/tests/unit/test_persist.py b/tests/unit/test_persist.py
index 025ad968..fafe6dc5 100644
--- a/tests/unit/test_persist.py
+++ b/tests/unit/test_persist.py
@@ -1,8 +1,10 @@
 import pytest
+from unittest.mock import patch
 
 from vcr.persisters.filesystem import FilesystemPersister
+from vcr.persisters.deduplicated_filesystem import DeduplicatedFilesystemPersister
 from vcr.serializers import jsonserializer, yamlserializer
-
+import vcr
 
 @pytest.mark.parametrize(
     "cassette_path, serializer",
@@ -28,3 +30,22 @@ def test_load_cassette_with_invalid_cassettes(cassette_path, serializer):
     with pytest.raises(Exception) as excinfo:
         FilesystemPersister.load_cassette(cassette_path, serializer)
     assert "run the migration script" not in excinfo.exconly()
+
+@pytest.mark.parametrize(
+    "cassette_path, serializer",
+    [
+        ("tests/fixtures/migration/cassette_with_duplicate_requests.yaml", yamlserializer),
+    ],
+)
+def test_load_cassette_with_duplicate_requests_cassettes(cassette_path, serializer):
+    cassette_dict = DeduplicatedFilesystemPersister.load_cassette(cassette_path, serializer)
+    breakpoint()
+    with patch.object(FilesystemPersister, "save_cassette") as mock:
+        with vcr.use_cassette(cassette_path, serializer=serializer, persister=DeduplicatedFilesystemPersister):
+            pass
+
+        # it's deduped when it is saved
+        # DeduplicatedFilesystemPersister.save_cassette(cassette_path, cassette_dict, serializer)
+        breakpoint()
+        assert mock.call_count == 1
+
diff --git a/vcr/cassette.py b/vcr/cassette.py
index 5822afac..c943bd07 100644
--- a/vcr/cassette.py
+++ b/vcr/cassette.py
@@ -12,6 +12,7 @@
 from .matchers import get_matchers_results, method, requests_match, uri
 from .patch import CassettePatcherBuilder
 from .persisters.filesystem import FilesystemPersister
+from .persisters.deduplicated_filesystem import DeduplicatedFilesystemPersister
 from .record_mode import RecordMode
 from .serializers import yamlserializer
 from .util import partition_dict
diff --git a/vcr/persisters/deduplicated_filesystem.py b/vcr/persisters/deduplicated_filesystem.py
new file mode 100644
index 00000000..2ba92b8d
--- /dev/null
+++ b/vcr/persisters/deduplicated_filesystem.py
@@ -0,0 +1,58 @@
+# https://github.com/Azure/azure-sdk-for-python/pull/17973/files
+
+import copy
+import os
+
+from vcr.serialize import serialize, deserialize
+from .filesystem import FilesystemPersister
+
+
+ATTRIBUTES_TO_COMPARE = [
+    "body",
+    "headers",
+    "host",
+    "method",
+    "path",
+    "protocol",
+    "query",
+    "scheme",
+    "uri",
+    "url",
+]
+
+
+def trim_duplicates(cassette_dict):
+    # Dict[str] -> Dict[str]
+    cassette_copy = copy.deepcopy(cassette_dict)
+    requests = cassette_dict["requests"]
+    responses = cassette_dict["responses"]
+    pairs_to_remove = []
+    for i in range(1, len(requests)):
+        for j in range(1, min(i, 4)):
+            if same_requests(requests[i - j], requests[i]):
+                pairs_to_remove.append(i - j)
+    # Always keep the last one
+    ret = {"requests": [], "responses": []}
+
+    for i in range(len(requests)):
+        if i not in pairs_to_remove:
+            ret["requests"].append(requests[i])
+            ret["responses"].append(responses[i])
+
+    return ret
+
+
+def same_requests(request1, request2):
+    # (vcr.Request, vcr.Request) -> bool
+    for attr in ATTRIBUTES_TO_COMPARE:
+        if getattr(request1, attr) != getattr(request2, attr):
+            return False
+
+    return True
+
+
+class DeduplicatedFilesystemPersister(FilesystemPersister):
+    @staticmethod
+    def save_cassette(cassette_path, cassette_dict, serializer):
+        cassette_dict = trim_duplicates(cassette_dict)
+        FilesystemPersister.save_cassette(cassette_path, cassette_dict, serializer)