diff --git a/glanceclient/common/http.py b/glanceclient/common/http.py
index b5bea8a9..a5fb1537 100644
--- a/glanceclient/common/http.py
+++ b/glanceclient/common/http.py
@@ -66,7 +66,7 @@ def encode_headers(headers):
     for h, v in headers.items():
         if v is not None:
             # if the item is token, do not quote '+' as well.
-            safe = '+/' if h in TOKEN_HEADERS else '/'
+            safe = '=+/' if h in TOKEN_HEADERS else '/'
             if six.PY2:
                 # incoming items may be unicode, so get them into something
                 # the py2 version of urllib can handle before percent encoding
diff --git a/glanceclient/tests/unit/test_http.py b/glanceclient/tests/unit/test_http.py
index cdc18953..6eee005c 100644
--- a/glanceclient/tests/unit/test_http.py
+++ b/glanceclient/tests/unit/test_http.py
@@ -467,11 +467,12 @@ class TestClient(testtools.TestCase):
         headers = self.mock.last_request.headers
         self.assertEqual(refreshed_token, headers['X-Auth-Token'])
         # regression check for bug 1448080
-        unicode_token = u'ni\xf1o+'
+        unicode_token = u'ni\xf1o+=='
         http_client.auth_token = unicode_token
         http_client.get(path)
         headers = self.mock.last_request.headers
         # Bug #1766235: According to RFC 8187, headers must be
         # encoded as 7-bit ASCII, so expect to see only displayable
-        # chars in percent-encoding. The '+' char will be not be changed.
-        self.assertEqual(b'ni%C3%B1o+', headers['X-Auth-Token'])
+        # chars in percent-encoding. The '+' and '= 'chars will not
+        # be changed.
+        self.assertEqual(b'ni%C3%B1o+==', headers['X-Auth-Token'])