diff --git a/keras_nlp/models/xlm_roberta/xlm_roberta_tokenizer.py b/keras_nlp/models/xlm_roberta/xlm_roberta_tokenizer.py
index d679ab9ba3..241ec4b5b6 100644
--- a/keras_nlp/models/xlm_roberta/xlm_roberta_tokenizer.py
+++ b/keras_nlp/models/xlm_roberta/xlm_roberta_tokenizer.py
@@ -123,10 +123,18 @@ def id_to_token(self, id):
 
     def token_to_id(self, token):
         """Convert a string token to an integer id."""
+
         if token in self._vocabulary_prefix:
             return self._vocabulary_prefix.index(token)
 
-        return int(self._sentence_piece.string_to_id(token).numpy()) + 1
+        spm_token_id = self._sentence_piece.string_to_id(token)
+
+        # OOV token
+        spm_unk_token_id = self._sentence_piece.string_to_id("<unk>")
+        if spm_token_id == spm_unk_token_id:
+            return self.unk_token_id
+
+        return int(spm_token_id.numpy()) + 1
 
     def tokenize(self, inputs):
         tokens = super().tokenize(inputs)
diff --git a/keras_nlp/models/xlm_roberta/xlm_roberta_tokenizer_test.py b/keras_nlp/models/xlm_roberta/xlm_roberta_tokenizer_test.py
index 912403fc4e..8ad1fddbfb 100644
--- a/keras_nlp/models/xlm_roberta/xlm_roberta_tokenizer_test.py
+++ b/keras_nlp/models/xlm_roberta/xlm_roberta_tokenizer_test.py
@@ -96,6 +96,10 @@ def test_id_to_token(self):
     def test_token_to_id(self):
         self.assertEqual(self.tokenizer.token_to_id("▁the"), 4)
         self.assertEqual(self.tokenizer.token_to_id("▁round"), 10)
+        # Test any random OOV token.
+        self.assertEqual(self.tokenizer.token_to_id("<oov-token>"), 3)
+        # Test a special token.
+        self.assertEqual(self.tokenizer.token_to_id("<pad>"), 1)
 
     def test_serialization(self):
         config = keras.utils.serialize_keras_object(self.tokenizer)