diff --git a/onnxoptimizer/test/function_test.py b/onnxoptimizer/test/function_test.py
new file mode 100644
index 000000000..77b1dc9b5
--- /dev/null
+++ b/onnxoptimizer/test/function_test.py
@@ -0,0 +1,34 @@
+import io
+import onnx
+import onnxoptimizer
+import pytest
+import unittest
+
+try:
+    import torch
+    import torchvision as tv
+
+    has_tv = True
+except:
+    has_tv = False
+
+
+@pytest.mark.skipif(not has_tv, reason="This test needs torchvision")
+def test_function_preserved():
+    with io.BytesIO() as f:
+        module = tv.models.resnet18()
+        torch.onnx.export(
+            module,
+            (torch.ones([1, 3, 224, 224], dtype=torch.float32), ),
+            f,
+            opset_version=15,
+            export_modules_as_functions={
+                torch.nn.BatchNorm2d,
+                torch.nn.Conv2d,
+            }
+        )
+
+        model = onnx.load_model_from_string(f.getvalue())
+        opt_model = onnxoptimizer.optimize(model)
+        assert len(model.functions) > 0
+        assert len(model.functions) == len(opt_model.functions)