diff --git a/Lib/test/test_xml_etree.py b/Lib/test/test_xml_etree.py
index 8a7ec0076ff036..e5a60ef80dc3e6 100644
--- a/Lib/test/test_xml_etree.py
+++ b/Lib/test/test_xml_etree.py
@@ -173,7 +173,7 @@ def assertEqualElements(self, alice, bob):
 # --------------------------------------------------------------------
 # element tree tests
 
-class ElementTreeTest(unittest.TestCase):
+class ElementTreeTest(ElementTestCase, unittest.TestCase):
 
     def serialize_check(self, elem, expected):
         self.assertEqual(serialize(elem), expected)
@@ -972,6 +972,79 @@ def test_qname(self):
         self.assertNotEqual(q1, 'ns:tag')
         self.assertEqual(q1, '{ns}tag')
 
+    def test_namespace_attribs(self):
+        # Unprefixed attributes are unqualified even if a default
+        # namespace is in effect. (This is a little unclear in some
+        # versions of the XML TR but is clarified in errata and other
+        # versions.) See bugs.python.org issue 17088.
+        #
+        # The reasoning behind this, alluded to in the spec, is that
+        # attribute meanings already depend on the element they're
+        # attached to; attributes have always lived in per-element
+        # namespaces even before explicit XML namespaces were
+        # introduced.  For that reason qualified attribute names are
+        # only really needed when one XML module defines attributes
+        # that can be placed on elements defined in a different module
+        # (such as happens with XLINK or, for that matter, the XML
+        # namespace spec itself).
+        e = ET.XML(
+            '<pf:elt xmlns:pf="space1" xmlns:pf2="space2" foo="value">'
+            '<pf:foo foo="value2" pf2:foo="value3"/>'
+            '<pf2:foo foo="value4" pf:foo="value5" pf2:foo="value6"/>'
+            '<foo foo="value7" pf:foo="value8"/>'
+            '</pf:elt>')
+        self.assertEqual(e.tag, '{space1}elt')
+        self.assertEqual(e.get('foo'), 'value')
+        self.assertIsNone(e.get('{space1}foo'))
+        self.assertIsNone(e.get('{space2}foo'))
+        self.assertEqual(e[0].tag, '{space1}foo')
+        self.assertEqual(e[0].attrib, { 'foo': 'value2',
+                                        '{space2}foo': 'value3' })
+        self.assertEqual(e[1].tag, '{space2}foo')
+        self.assertEqual(e[1].attrib, { 'foo': 'value4',
+                                        '{space1}foo': 'value5',
+                                        '{space2}foo': 'value6' })
+        self.assertEqual(e[2].tag, 'foo')
+        self.assertEqual(e[2].attrib, { 'foo': 'value7',
+                                        '{space1}foo': 'value8' })
+
+        serialized1 = (
+            '<ns0:elt xmlns:ns0="space1" xmlns:ns1="space2" foo="value">'
+            '<ns0:foo foo="value2" ns1:foo="value3" />'
+            '<ns1:foo foo="value4" ns0:foo="value5" ns1:foo="value6" />'
+            '<foo foo="value7" ns0:foo="value8" />'
+            '</ns0:elt>')
+        self.assertEqual(serialize(e), serialized1)
+        self.assertEqualElements(e, ET.XML(serialized1))
+
+        # Test writing with a default namespace.
+        with self.assertRaisesRegex(ValueError,
+                'cannot use non-qualified name.* with default_namespace option'):
+            serialize(e, default_namespace="space1")
+
+        # Remove the unqualified element from the tree so we can test
+        # further.
+        del e[2]
+
+        # Serialization can require a namespace prefix to be declared for
+        # space1 even if no elements use that prefix, in order to
+        # write an attribute name in that namespace.
+        serialized2 = (
+            '<ns0:elt xmlns="space2" xmlns:ns0="space1" xmlns:ns1="space2" foo="value">'
+            '<ns0:foo foo="value2" ns1:foo="value3" />'
+            '<foo foo="value4" ns0:foo="value5" ns1:foo="value6" />'
+            '</ns0:elt>')
+        self.assertEqual(serialize(e, default_namespace="space2"), serialized2)
+        self.assertEqualElements(e, ET.XML(serialized2))
+
+        serialized3 = (
+            '<elt xmlns="space1" xmlns:ns0="space2" xmlns:ns1="space1" foo="value">'
+            '<foo foo="value2" ns0:foo="value3" />'
+            '<ns0:foo foo="value4" ns1:foo="value5" ns0:foo="value6" />'
+            '</elt>')
+        self.assertEqual(serialize(e, default_namespace="space1"), serialized3)
+        self.assertEqualElements(e, ET.XML(serialized3))
+
     def test_doctype_public(self):
         # Test PUBLIC doctype.
 
@@ -1619,18 +1692,17 @@ def test_bug_200709_default_namespace(self):
         s = ET.SubElement(e, "{default}elem")
         s = ET.SubElement(e, "{not-default}elem")
         self.assertEqual(serialize(e, default_namespace="default"), # 2
-            '<elem xmlns="default" xmlns:ns1="not-default">'
+            '<elem xmlns="default" xmlns:ns0="not-default">'
             '<elem />'
-            '<ns1:elem />'
+            '<ns0:elem />'
             '</elem>')
 
         e = ET.Element("{default}elem")
         s = ET.SubElement(e, "{default}elem")
         s = ET.SubElement(e, "elem") # unprefixed name
-        with self.assertRaises(ValueError) as cm:
+        with self.assertRaisesRegex(ValueError,
+                'cannot use non-qualified name.* with default_namespace option'):
             serialize(e, default_namespace="default") # 3
-        self.assertEqual(str(cm.exception),
-                'cannot use non-qualified names with default_namespace option')
 
     def test_bug_200709_register_namespace(self):
         e = ET.Element("{http://namespace.invalid/does/not/exist/}title")
diff --git a/Lib/xml/etree/ElementTree.py b/Lib/xml/etree/ElementTree.py
index c1cf483cf56bb2..0683020352de97 100644
--- a/Lib/xml/etree/ElementTree.py
+++ b/Lib/xml/etree/ElementTree.py
@@ -836,19 +836,21 @@ def _get_writer(file_or_filename, encoding):
 def _namespaces(elem, default_namespace=None):
     # identify namespaces used in this tree
 
-    # maps qnames to *encoded* prefix:local names
-    qnames = {None: None}
+    # maps qnames to *encoded* prefix:local names,
+    # value is a pair of prefix:local strings,
+    # second value is for attribute names
+    qnames = {None: (None, None)}
 
     # maps uri:s to prefixes
     namespaces = {}
-    if default_namespace:
-        namespaces[default_namespace] = ""
 
-    def add_qname(qname):
+    def serialize_qname(qname, is_attr):
         # calculate serialized qname representation
-        try:
-            if qname[:1] == "{":
-                uri, tag = qname[1:].rsplit("}", 1)
+        if qname[:1] == "{":
+            uri, local = qname[1:].rsplit("}", 1)
+            if not is_attr and uri == default_namespace:
+                prefix = ""
+            else:
                 prefix = namespaces.get(uri)
                 if prefix is None:
                     prefix = _namespace_map.get(uri)
@@ -856,43 +858,59 @@ def add_qname(qname):
                         prefix = "ns%d" % len(namespaces)
                     if prefix != "xml":
                         namespaces[uri] = prefix
-                if prefix:
-                    qnames[qname] = "%s:%s" % (prefix, tag)
-                else:
-                    qnames[qname] = tag # default element
+            if prefix:
+                return "%s:%s" % (prefix, local)
             else:
-                if default_namespace:
-                    # FIXME: can this be handled in XML 1.0?
-                    raise ValueError(
-                        "cannot use non-qualified names with "
-                        "default_namespace option"
-                        )
-                qnames[qname] = qname
-        except TypeError:
-            _raise_serialization_error(qname)
+                return local # default element
+        else:
+            if not is_attr and default_namespace:
+                # FIXME: can this be handled in XML 1.0?
+                raise ValueError(
+                    "cannot use non-qualified name (<%s>) with "
+                    "default_namespace option" % qname
+                    )
+            return qname
+
+    def add_qname(qname, is_attr=False):
+        ser_tag, ser_attr = qnames.get(qname, (None, None))
+        if is_attr:
+            if ser_attr is None:
+                ser_attr = serialize_qname(qname, True)
+                if not default_namespace:
+                    ser_tag = ser_attr
+                qnames[qname] = (ser_tag, ser_attr)
+        else:
+            if ser_tag is None:
+                ser_tag = serialize_qname(qname, False)
+                if not default_namespace:
+                    ser_attr = ser_tag
+                qnames[qname] = (ser_tag, ser_attr)
 
     # populate qname and namespaces table
     for elem in elem.iter():
         tag = elem.tag
         if isinstance(tag, QName):
-            if tag.text not in qnames:
-                add_qname(tag.text)
+            add_qname(tag.text)
         elif isinstance(tag, str):
-            if tag not in qnames:
-                add_qname(tag)
+            add_qname(tag)
         elif tag is not None and tag is not Comment and tag is not PI:
             _raise_serialization_error(tag)
         for key, value in elem.items():
             if isinstance(key, QName):
                 key = key.text
-            if key not in qnames:
-                add_qname(key)
-            if isinstance(value, QName) and value.text not in qnames:
+            elif not isinstance(key, str):
+                _raise_serialization_error(key)
+            add_qname(key, is_attr=True)
+            if isinstance(value, QName):
                 add_qname(value.text)
         text = elem.text
-        if isinstance(text, QName) and text.text not in qnames:
+        if isinstance(text, QName):
             add_qname(text.text)
-    return qnames, namespaces
+
+    prefix_map = {prefix: ns for ns, prefix in namespaces.items()}
+    if default_namespace:
+        prefix_map[""] = default_namespace
+    return qnames, prefix_map
 
 def _serialize_xml(write, elem, qnames, namespaces,
                    short_empty_elements, **kwargs):
@@ -903,7 +921,7 @@ def _serialize_xml(write, elem, qnames, namespaces,
     elif tag is ProcessingInstruction:
         write("<?%s?>" % text)
     else:
-        tag = qnames[tag]
+        tag = qnames[tag][0]
         if tag is None:
             if text:
                 write(_escape_cdata(text))
@@ -915,8 +933,7 @@ def _serialize_xml(write, elem, qnames, namespaces,
             items = list(elem.items())
             if items or namespaces:
                 if namespaces:
-                    for v, k in sorted(namespaces.items(),
-                                       key=lambda x: x[1]):  # sort on prefix
+                    for k, v in sorted(namespaces.items()):
                         if k:
                             k = ":" + k
                         write(" xmlns%s=\"%s\"" % (
@@ -927,10 +944,10 @@ def _serialize_xml(write, elem, qnames, namespaces,
                     if isinstance(k, QName):
                         k = k.text
                     if isinstance(v, QName):
-                        v = qnames[v.text]
+                        v = qnames[v.text][0]
                     else:
                         v = _escape_attrib(v)
-                    write(" %s=\"%s\"" % (qnames[k], v))
+                    write(" %s=\"%s\"" % (qnames[k][1], v))
             if text or len(elem) or not short_empty_elements:
                 write(">")
                 if text:
@@ -960,7 +977,7 @@ def _serialize_html(write, elem, qnames, namespaces, **kwargs):
     elif tag is ProcessingInstruction:
         write("<?%s?>" % _escape_cdata(text))
     else:
-        tag = qnames[tag]
+        tag = qnames[tag][0]
         if tag is None:
             if text:
                 write(_escape_cdata(text))
@@ -971,8 +988,7 @@ def _serialize_html(write, elem, qnames, namespaces, **kwargs):
             items = list(elem.items())
             if items or namespaces:
                 if namespaces:
-                    for v, k in sorted(namespaces.items(),
-                                       key=lambda x: x[1]):  # sort on prefix
+                    for k, v in sorted(namespaces.items()):
                         if k:
                             k = ":" + k
                         write(" xmlns%s=\"%s\"" % (
@@ -983,11 +999,11 @@ def _serialize_html(write, elem, qnames, namespaces, **kwargs):
                     if isinstance(k, QName):
                         k = k.text
                     if isinstance(v, QName):
-                        v = qnames[v.text]
+                        v = qnames[v.text][0]
                     else:
                         v = _escape_attrib_html(v)
                     # FIXME: handle boolean attributes
-                    write(" %s=\"%s\"" % (qnames[k], v))
+                    write(" %s=\"%s\"" % (qnames[k][1], v))
             write(">")
             ltag = tag.lower()
             if text:
diff --git a/Misc/NEWS.d/next/Library/2018-12-09-14-58-08.bpo-17088.AHFvrn.rst b/Misc/NEWS.d/next/Library/2018-12-09-14-58-08.bpo-17088.AHFvrn.rst
new file mode 100644
index 00000000000000..e92d690d4331fd
--- /dev/null
+++ b/Misc/NEWS.d/next/Library/2018-12-09-14-58-08.bpo-17088.AHFvrn.rst
@@ -0,0 +1,5 @@
+ElementTree's serialization of attribute names when a default namespace is
+passed was corrected. Previously, unqualified attribute names would be
+rejected unnecessarily, while attributes in the default namespace would have
+their prefix stripped, which goes against the rules for namespace defaulting
+and uniqueness of attributes as specified in the XML namespaces spec.