diff --git a/Lib/fontTools/ttLib/tables/_n_a_m_e.py b/Lib/fontTools/ttLib/tables/_n_a_m_e.py
index 2383f7953..e9ff2151d 100644
--- a/Lib/fontTools/ttLib/tables/_n_a_m_e.py
+++ b/Lib/fontTools/ttLib/tables/_n_a_m_e.py
@@ -459,22 +459,32 @@ class NameRecord(object):
 		if type(self) != type(other):
 			return NotImplemented
 
-		# implemented so that list.sort() sorts according to the spec.
-		selfTuple = (
-			getattr(self, "platformID", None),
-			getattr(self, "platEncID", None),
-			getattr(self, "langID", None),
-			getattr(self, "nameID", None),
-			getattr(self, "string", None),
-		)
-		otherTuple = (
-			getattr(other, "platformID", None),
-			getattr(other, "platEncID", None),
-			getattr(other, "langID", None),
-			getattr(other, "nameID", None),
-			getattr(other, "string", None),
-		)
-		return selfTuple < otherTuple
+		try:
+			# implemented so that list.sort() sorts according to the spec.
+			selfTuple = (
+				self.platformID,
+				self.platEncID,
+				self.langID,
+				self.nameID,
+				self.toBytes(),
+			)
+			otherTuple = (
+				other.platformID,
+				other.platEncID,
+				other.langID,
+				other.nameID,
+				other.toBytes(),
+			)
+			return selfTuple < otherTuple
+		except (UnicodeEncodeError, AttributeError):
+			# This can only happen for
+			# 1) an object that is not a NameRecord, or
+			# 2) an unlikely incomplete NameRecord object which has not been
+			#    fully populated, or
+			# 3) when all IDs are identical but the strings can't be encoded
+			#    for their platform encoding.
+			# In all cases it is best to return NotImplemented.
+			return NotImplemented
 
 	def __repr__(self):
 		return "<NameRecord NameID=%d; PlatformID=%d; LanguageID=%d>" % (
diff --git a/Tests/ttLib/tables/_n_a_m_e_test.py b/Tests/ttLib/tables/_n_a_m_e_test.py
index 98047e452..d770a5231 100644
--- a/Tests/ttLib/tables/_n_a_m_e_test.py
+++ b/Tests/ttLib/tables/_n_a_m_e_test.py
@@ -53,6 +53,26 @@ class NameTableTest(unittest.TestCase):
 		with self.assertRaises(TypeError):
 			table.setName(1.000, 5, 1, 0, 0)
 
+	def test_names_sort_bytes_str(self):
+		# Corner case: If a user appends a name record directly to `names`, the
+		# `__lt__` method on NameRecord may run into duplicate name records where
+		# one `string` is a str and the other one bytes, leading to an exception.
+		table = table__n_a_m_e()
+		table.names = [
+			makeName("Test", 25, 3, 1, 0x409),
+			makeName("Test".encode("utf-16be"), 25, 3, 1, 0x409),
+		]
+		table.compile(None)
+
+	def test_names_sort_bytes_str_encoding_error(self):
+		table = table__n_a_m_e()
+		table.names = [
+			makeName("Test寬", 25, 1, 0, 0),
+			makeName("Test鬆鬆", 25, 1, 0, 0),
+		]
+		with self.assertRaises(TypeError):
+			table.names.sort()
+
 	def test_addName(self):
 		table = table__n_a_m_e()
 		nameIDs = []