diff --git a/Lib/email/_encoded_words.py b/Lib/email/_encoded_words.py
index c40ffa9..295ae7e 100644
--- a/Lib/email/_encoded_words.py
+++ b/Lib/email/_encoded_words.py
@@ -98,30 +98,42 @@
 #
 
 def decode_b(encoded):
-    defects = []
+    # First try encoding with validate=True, fixing the padding if needed.
+    # This will succeed only if encoded includes no invalid characters.
     pad_err = len(encoded) % 4
-    if pad_err:
-        defects.append(errors.InvalidBase64PaddingDefect())
-        padded_encoded = encoded + b'==='[:4-pad_err]
-    else:
-        padded_encoded = encoded
+    missing_padding = b'==='[:4-pad_err] if pad_err else b''
     try:
-        return base64.b64decode(padded_encoded, validate=True), defects
+        return (
+            base64.b64decode(encoded + missing_padding, validate=True),
+            [errors.InvalidBase64PaddingDefect()] if pad_err else [],
+        )
     except binascii.Error:
-        # Since we had correct padding, this must an invalid char error.
-        defects = [errors.InvalidBase64CharactersDefect()]
+        # Since we had correct padding, this is likely an invalid char error.
+        #
         # The non-alphabet characters are ignored as far as padding
-        # goes, but we don't know how many there are.  So we'll just
-        # try various padding lengths until something works.
-        for i in 0, 1, 2, 3:
+        # goes, but we don't know how many there are.  So try without adding
+        # padding to see if it works.
+        try:
+            return (
+                base64.b64decode(encoded, validate=False),
+                [errors.InvalidBase64CharactersDefect()],
+            )
+        except binascii.Error:
+            # Add as much padding as could possibly be necessary (extra padding
+            # is ignored).
             try:
-                return base64.b64decode(encoded+b'='*i, validate=False), defects
+                return (
+                    base64.b64decode(encoded + b'==', validate=False),
+                    [errors.InvalidBase64CharactersDefect(),
+                     errors.InvalidBase64PaddingDefect()],
+                )
             except binascii.Error:
-                if i==0:
-                    defects.append(errors.InvalidBase64PaddingDefect())
-        else:
-            # This should never happen.
-            raise AssertionError("unexpected binascii.Error")
+                # This only happens when the encoded string's length is 1 more
+                # than a multiple of 4, which is invalid.
+                #
+                # bpo-27397: Just return the encoded string since there's no
+                # way to decode.
+                return encoded, [errors.InvalidBase64LengthDefect()]
 
 def encode_b(bstring):
     return base64.b64encode(bstring).decode('ascii')
diff --git a/Lib/email/errors.py b/Lib/email/errors.py
index 791239f..d28a680 100644
--- a/Lib/email/errors.py
+++ b/Lib/email/errors.py
@@ -73,6 +73,9 @@
 class InvalidBase64CharactersDefect(MessageDefect):
     """base64 encoded sequence had characters not in base64 alphabet"""
 
+class InvalidBase64LengthDefect(MessageDefect):
+    """base64 encoded sequence had invalid length (1 mod 4)"""
+
 # These errors are specific to header parsing.
 
 class HeaderDefect(MessageDefect):
diff --git a/Lib/test/test_email/test__encoded_words.py b/Lib/test/test_email/test__encoded_words.py
index 900e1d0..5a59aeb 100644
--- a/Lib/test/test_email/test__encoded_words.py
+++ b/Lib/test/test_email/test__encoded_words.py
@@ -33,7 +33,10 @@
         self._test(b'Zm9v', b'foo')
 
     def test_missing_padding(self):
+        # 1 missing padding character
         self._test(b'dmk', b'vi', [errors.InvalidBase64PaddingDefect])
+        # 2 missing padding characters
+        self._test(b'dg', b'v', [errors.InvalidBase64PaddingDefect])
 
     def test_invalid_character(self):
         self._test(b'dm\x01k===', b'vi', [errors.InvalidBase64CharactersDefect])
@@ -42,6 +45,9 @@
         self._test(b'dm\x01k', b'vi', [errors.InvalidBase64CharactersDefect,
                                        errors.InvalidBase64PaddingDefect])
 
+    def test_invalid_length(self):
+        self._test(b'abcde', b'abcde', [errors.InvalidBase64LengthDefect])
+
 
 class TestDecode(TestEmailBase):
 
diff --git a/Lib/test/test_email/test__header_value_parser.py b/Lib/test/test_email/test__header_value_parser.py
index 5cdc4bc..5036de2 100644
--- a/Lib/test/test_email/test__header_value_parser.py
+++ b/Lib/test/test_email/test__header_value_parser.py
@@ -347,6 +347,15 @@
              errors.InvalidBase64PaddingDefect],
             '')
 
+    def test_get_unstructured_invalid_base64_length(self):
+        # bpo-27397: Return the encoded string since there's no way to decode.
+        self._test_get_x(self._get_unst,
+            '=?utf-8?b?abcde?=',
+            'abcde',
+            'abcde',
+            [errors.InvalidBase64LengthDefect],
+            '')
+
     def test_get_unstructured_no_whitespace_between_ews(self):
         self._test_get_x(self._get_unst,
             '=?utf-8?q?foo?==?utf-8?q?bar?=',
diff --git a/Lib/test/test_email/test_defect_handling.py b/Lib/test/test_email/test_defect_handling.py
index f36b907..781f657 100644
--- a/Lib/test/test_email/test_defect_handling.py
+++ b/Lib/test/test_email/test_defect_handling.py
@@ -254,6 +254,23 @@
         self.assertDefectsEqual(self.get_defects(msg),
                                 [errors.InvalidBase64CharactersDefect])
 
+    def test_invalid_length_of_base64_payload(self):
+        source = textwrap.dedent("""\
+            Subject: test
+            MIME-Version: 1.0
+            Content-Type: text/plain; charset="utf-8"
+            Content-Transfer-Encoding: base64
+
+            abcde
+            """)
+        msg = self._str_msg(source)
+        with self._raise_point(errors.InvalidBase64LengthDefect):
+            payload = msg.get_payload(decode=True)
+        if self.raise_expected: return
+        self.assertEqual(payload, b'abcde')
+        self.assertDefectsEqual(self.get_defects(msg),
+                                [errors.InvalidBase64LengthDefect])
+
     def test_missing_ending_boundary(self):
         source = textwrap.dedent("""\
             To: 1@harrydomain4.com