diff --git a/opencompass/datasets/tydiqa.py b/opencompass/datasets/tydiqa.py
index c27e738d9..1398b1943 100644
--- a/opencompass/datasets/tydiqa.py
+++ b/opencompass/datasets/tydiqa.py
@@ -75,6 +75,7 @@ def score(self, predictions, references):
             }
         for prediction, reference in zip(predictions, references):
             prediction = re.split(r'[\n]', prediction, 1)[0].lower()
+            reference = [ref.lower() for ref in reference]
             exact_match += self.metric_max_over_ground_truths(
                 self.exact_match_score, prediction, reference)
             f1 += self.metric_max_over_ground_truths(self.f1_score, prediction,
diff --git a/tests/datasets/test_tydiqa.py b/tests/datasets/test_tydiqa.py
new file mode 100644
index 000000000..1dcaa8f21
--- /dev/null
+++ b/tests/datasets/test_tydiqa.py
@@ -0,0 +1,24 @@
+import unittest
+
+from opencompass.datasets.tydiqa import TydiQAEvaluator
+
+
+class TestTydiQAEvaluator(unittest.TestCase):
+
+    def test_score_ignores_case_of_references(self):
+        evaluator = TydiQAEvaluator()
+        result = evaluator.score(
+            predictions=['Paris', 'Barack Obama was president'],
+            references=[['Paris'], ['Barack Obama']])
+        self.assertEqual(result['exact_match'], 50.0)
+        self.assertAlmostEqual(result['f1'], (100.0 + 200 / 3) / 2)
+
+    def test_score_wrong_answer(self):
+        evaluator = TydiQAEvaluator()
+        result = evaluator.score(predictions=['London'],
+                                 references=[['Paris']])
+        self.assertEqual(result, {'exact_match': 0.0, 'f1': 0.0})
+
+
+if __name__ == '__main__':
+    unittest.main()