diff --git a/doc/howto/dev/new_op_cn.md b/doc/howto/dev/new_op_cn.md
index 7f8da2da5a0d42ff065265c5d173d0e6167dc08a..3e71a0a592861b32861c0e402911e274208fe714 100644
--- a/doc/howto/dev/new_op_cn.md
+++ b/doc/howto/dev/new_op_cn.md
@@ -227,6 +227,12 @@ make mul_op
     USE_CPU_ONLY_OP(gather);
     ```
     
+    如果OP不带Kernel，则使用`USE_NO_KENREL_OP`:
+    
+    ```
+    USE_NO_KENREL_OP(recurrent);
+    ```
+    
     使用`USE_OP`告知编译器需要链接该Op的目标文件，具体解释参考[代码注释](https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/framework/op_registry.h#L81)。
     
     
diff --git a/paddle/framework/backward.cc b/paddle/framework/backward.cc
index 6b4c612cd8d9263258e3987914c44002e7bca92c..c5d46622156c56acb98fb77e7db5ee7bca8c937a 100644
--- a/paddle/framework/backward.cc
+++ b/paddle/framework/backward.cc
@@ -182,7 +182,7 @@ static std::unique_ptr<OperatorBase> BackwardRecursive(
                    });
 
     // process recurrent gradient op as a special operator.
-    if (forwardOp.Type() == "recurrent_op") {
+    if (forwardOp.Type() == "recurrent") {
       // NOTE clean up cycle call somewhere (RNN's stepnet constains itself), or
       // this will result in infinite loop.
       const auto& rnnop =
diff --git a/paddle/framework/op_registry.h b/paddle/framework/op_registry.h
index 2d09cde41e3f5086279f9441e0fdc52549bed5ab..572dff860a306bb03ba9e6702fec85e4a2ea1b54 100644
--- a/paddle/framework/op_registry.h
+++ b/paddle/framework/op_registry.h
@@ -199,6 +199,8 @@ class OpKernelRegistrar : public Registrar {
   USE_OP_DEVICE_KERNEL(op_type, GPU)
 #endif
 
+#define USE_NO_KERNEL_OP(op_type) USE_OP_ITSELF(op_type);
+
 #define USE_CPU_ONLY_OP(op_type) \
   USE_OP_ITSELF(op_type);        \
   USE_OP_DEVICE_KERNEL(op_type, CPU);
diff --git a/paddle/operators/minus_op.cc b/paddle/operators/minus_op.cc
index 1eee9644babbdfac68821ca774845ad8ebbd5aee..069fb5e1abc657aa02a50fde352ce88d078c36e1 100644
--- a/paddle/operators/minus_op.cc
+++ b/paddle/operators/minus_op.cc
@@ -79,7 +79,7 @@ class MinusGradOp : public NetOp {
 }  // namespace paddle
 
 USE_OP(scale);
-USE_OP_ITSELF(identity);
+USE_NO_KERNEL_OP(identity);
 namespace ops = paddle::operators;
 REGISTER_OP(minus, ops::MinusOp, ops::MinusOpMaker, minus_grad,
             ops::MinusGradOp<float>);
diff --git a/paddle/operators/recurrent_op.cc b/paddle/operators/recurrent_op.cc
index 16bd249cb3d989c695ec9378f09d48833d70be58..e826703c60ca82e1fe690eb78c3d4f92981ef3a2 100644
--- a/paddle/operators/recurrent_op.cc
+++ b/paddle/operators/recurrent_op.cc
@@ -235,5 +235,5 @@ RecurrentGradientOp::RecurrentGradientOp(
 }  // namespace paddle
 
 REGISTER_OP_WITHOUT_GRADIENT(
-    recurrent_op, paddle::operators::RecurrentOp,
+    recurrent, paddle::operators::RecurrentOp,
     paddle::operators::RecurrentAlgorithmProtoAndCheckerMaker);
diff --git a/paddle/pybind/pybind.cc b/paddle/pybind/pybind.cc
index 3bc150ccb7af2885439cc2344aa0db9ba3b1ca03..6896422617be0a3c73dc7b0d7cc1113075fa2f4b 100644
--- a/paddle/pybind/pybind.cc
+++ b/paddle/pybind/pybind.cc
@@ -39,12 +39,12 @@ USE_OP(sigmoid);
 USE_OP(softmax);
 USE_OP(rowwise_add);
 USE_OP(fill_zeros_like);
-USE_OP_ITSELF(recurrent_op);
+USE_NO_KERNEL_OP(recurrent);
 USE_OP(gaussian_random);
 USE_OP(uniform_random);
 USE_OP(lookup_table);
 USE_OP(scale);
-USE_OP_ITSELF(identity);
+USE_NO_KERNEL_OP(identity);
 USE_OP(minus);
 USE_CPU_ONLY_OP(gather);
 USE_CPU_ONLY_OP(scatter);
diff --git a/python/paddle/v2/framework/op.py b/python/paddle/v2/framework/op.py
index 6ac656321e72f5b0c91008091753ee50ac8200a6..e7e932f6fe46f2db8de37cecfd9875b310bdcbe5 100644
--- a/python/paddle/v2/framework/op.py
+++ b/python/paddle/v2/framework/op.py
@@ -179,7 +179,7 @@ class OperatorFactory(object):
 
 class __RecurrentOp__(object):
     __proto__ = None
-    type = 'recurrent_op'
+    type = 'recurrent'
 
     def __init__(self):
         # cache recurrent_op's proto