diff --git a/python/paddle/fluid/tests/unittests/test_sync_batch_norm_op.py b/python/paddle/fluid/tests/unittests/test_sync_batch_norm_op.py
index 4fa64bef32fff0312e5076e854026226754e5b1c..baac0af5d61afe570f46cd65d292262f98e219af 100644
--- a/python/paddle/fluid/tests/unittests/test_sync_batch_norm_op.py
+++ b/python/paddle/fluid/tests/unittests/test_sync_batch_norm_op.py
@@ -25,6 +25,7 @@ import six
 import paddle
 import paddle.fluid.core as core
 import paddle.fluid as fluid
+import paddle.nn as nn
 from paddle.fluid import compiler
 from paddle.fluid import Program, program_guard
 
@@ -244,5 +245,34 @@ class TestConvertSyncBatchNorm(unittest.TestCase):
                         isinstance(model[idx], paddle.nn.SyncBatchNorm), True)
 
 
+class TestConvertSyncBatchNormCase2(unittest.TestCase):
+    def test_convert(self):
+        if not core.is_compiled_with_cuda():
+            return
+
+        class Net(nn.Layer):
+            def __init__(self):
+                super(Net, self).__init__()
+                self.conv1 = nn.Conv2D(3, 5, 3)
+                self.bn = []
+                bn = self.add_sublayer('bn', nn.BatchNorm2D(5))
+                self.bn.append(bn)
+
+            def forward(self, x):
+                x = self.conv1(x)
+                for bn in self.bn:
+                    x = bn(x)
+                return x
+
+        model = nn.Sequential()
+        model.add_sublayer('net1', Net())
+        model.add_sublayer('net2', Net())
+        compare_model = nn.Sequential()
+        compare_model.add_sublayer('net1', Net())
+        compare_model.add_sublayer('net2', Net())
+        model = nn.SyncBatchNorm.convert_sync_batchnorm(model)
+        self.assertEqual(len(compare_model.sublayers()), len(model.sublayers()))
+
+
 if __name__ == '__main__':
     unittest.main()
diff --git a/python/paddle/nn/layer/norm.py b/python/paddle/nn/layer/norm.py
index d8a4066cf0311e8425c537585d73434709eabd63..256694c7af67dd23a5056d929b8788676cddcb56 100644
--- a/python/paddle/nn/layer/norm.py
+++ b/python/paddle/nn/layer/norm.py
@@ -1121,7 +1121,7 @@ class SyncBatchNorm(_BatchNormBase):
             layer_output._mean = layer._mean
             layer_output._variance = layer._variance
 
-        for name, sublayer in layer.named_sublayers():
+        for name, sublayer in layer.named_children():
             layer_output.add_sublayer(name,
                                       cls.convert_sync_batchnorm(sublayer))
         del layer