diff --git a/tests/integration/test_requests.py b/tests/integration/test_requests.py
index e309d3f..235b6b1 100644
--- a/tests/integration/test_requests.py
+++ b/tests/integration/test_requests.py
@@ -167,20 +167,40 @@ def test_cross_scheme(tmpdir, httpbin_secure, httpbin):
         assert len(cass) == 2
 
 
-def test_gzip(tmpdir, httpbin_both):
+def test_gzip__decode_compressed_response_false(tmpdir, httpbin_both):
     """
     Ensure that requests (actually urllib3) is able to automatically decompress
     the response body
     """
+    for _ in range(2):  # one for recording, one for re-playing
+        with vcr.use_cassette(str(tmpdir.join("gzip.yaml"))):
+            response = requests.get(httpbin_both + "/gzip")
+            assert response.headers["content-encoding"] == "gzip"  # i.e. not removed
+            assert_is_json(response.content)  # i.e. uncompressed bytes
+
+
+def test_gzip__decode_compressed_response_true(tmpdir, httpbin_both):
     url = httpbin_both + "/gzip"
-    response = requests.get(url)
 
-    with vcr.use_cassette(str(tmpdir.join("gzip.yaml"))):
-        response = requests.get(url)
-        assert_is_json(response.content)
+    expected_response = requests.get(url)
+    expected_content = expected_response.content
+    assert expected_response.headers["content-encoding"] == "gzip"  # self-test
 
-    with vcr.use_cassette(str(tmpdir.join("gzip.yaml"))):
-        assert_is_json(response.content)
+    with vcr.use_cassette(
+        str(tmpdir.join("decode_compressed.yaml")), decode_compressed_response=True
+    ) as cassette:
+        r = requests.get(url)
+        assert r.headers["content-encoding"] == "gzip"  # i.e. not removed
+        assert r.content == expected_content
+
+    # Has the cassette body been decompressed?
+    cassette_response_body = cassette.responses[0]["body"]["string"]
+    assert isinstance(cassette_response_body, str)
+
+    with vcr.use_cassette(str(tmpdir.join("decode_compressed.yaml")), decode_compressed_response=True):
+        r = requests.get(url)
+        assert "content-encoding" not in r.headers  # i.e. removed
+        assert r.content == expected_content
 
 
 def test_session_and_connection_close(tmpdir, httpbin):
diff --git a/tests/unit/test_filters.py b/tests/unit/test_filters.py
index 370699d..003e08e 100644
--- a/tests/unit/test_filters.py
+++ b/tests/unit/test_filters.py
@@ -298,6 +298,18 @@ def test_decode_response_deflate():
     assert decoded_response["headers"]["content-length"] == [str(len(body))]
 
 
+def test_decode_response_deflate_already_decompressed():
+    body = b"deflate message"
+    gzip_response = {
+        "body": {"string": body},
+        "headers": {
+            "content-encoding": ["deflate"],
+        },
+    }
+    decoded_response = decode_response(gzip_response)
+    assert decoded_response["body"]["string"] == body
+
+
 def test_decode_response_gzip():
     body = b"gzip message"
 
@@ -325,3 +337,15 @@ def test_decode_response_gzip():
     decoded_response = decode_response(gzip_response)
     assert decoded_response["body"]["string"] == body
     assert decoded_response["headers"]["content-length"] == [str(len(body))]
+
+
+def test_decode_response_gzip_already_decompressed():
+    body = b"gzip message"
+    gzip_response = {
+        "body": {"string": body},
+        "headers": {
+            "content-encoding": ["gzip"],
+        },
+    }
+    decoded_response = decode_response(gzip_response)
+    assert decoded_response["body"]["string"] == body
diff --git a/vcr/filters.py b/vcr/filters.py
index 751c5d4..ac07fb0 100644
--- a/vcr/filters.py
+++ b/vcr/filters.py
@@ -153,9 +153,15 @@ def decode_response(response):
         if not body:
             return ""
         if encoding == "gzip":
-            return zlib.decompress(body, zlib.MAX_WBITS | 16)
+            try:
+                return zlib.decompress(body, zlib.MAX_WBITS | 16)
+            except zlib.error:
+                return body  # assumes that the data was already decompressed
         else:  # encoding == 'deflate'
-            return zlib.decompress(body)
+            try:
+                return zlib.decompress(body)
+            except zlib.error:
+                return body  # assumes that the data was already decompressed
 
     # Deepcopy here in case `headers` contain objects that could
     # be mutated by a shallow copy and corrupt the real response.