diff --git a/bindings/python/cntk/tests/onnx_op_test.py b/bindings/python/cntk/tests/onnx_op_test.py
index ad4503e0b8df..ff3f0bc544c8 100644
--- a/bindings/python/cntk/tests/onnx_op_test.py
+++ b/bindings/python/cntk/tests/onnx_op_test.py
@@ -426,9 +426,8 @@ def test_AveragePool(tmpdir, dtype, device_id):
 #AveragePool
 @pytest.mark.parametrize("dtype", DType_Config)
 def test_AveragePoolWithSequenceAxis(tmpdir, dtype, device_id):
-    if dtype == np.float16:
-        # CI reporting "FP16 convolution is only supported via cuDNN." on GPU with float16 data
-        pytest.skip('Test is skipped with float16 data')
+    if device_id == -1 and dtype == np.float16:
+        pytest.skip('Test is skipped on CPU with float16 data')
     device = cntk_device(device_id)
     with C.default_options(dtype=dtype):
         img = np.reshape(np.arange(16, dtype = dtype), [1, 4, 4])
@@ -1343,9 +1342,8 @@ def test_MaxPool(tmpdir, dtype, device_id):
 #MaxPool
 @pytest.mark.parametrize("dtype", DType_Config)
 def test_MaxPoolWithSequenceAxis(tmpdir, dtype, device_id):
-    if dtype == np.float16:
-        # CI reporting "FP16 convolution is only supported via cuDNN." on GPU with float16 data
-        pytest.skip('Test is skipped with float16 data')
+    if device_id == -1 and dtype == np.float16:
+        pytest.skip('Test is skipped on CPU with float16 data')
     device = cntk_device(device_id)
     with C.default_options(dtype=dtype):
         img = np.reshape(np.arange(16, dtype = dtype), [1, 4, 4])