diff --git a/python/paddle/fluid/layers/ops.py b/python/paddle/fluid/layers/ops.py
index 37784805f02f63d259a7d25e99d89c82dd6880a3..072b95969c7bea272e38e00d0ff6be7cd9ef3a04 100644
--- a/python/paddle/fluid/layers/ops.py
+++ b/python/paddle/fluid/layers/ops.py
@@ -239,3 +239,82 @@ Examples:
         # array([[ 0.21134382, -0.        ,  0.32876605],
         #        [-0.        , -0.        ,  1.0013918 ]], dtype=float32)
 """
+
+__all__ += ['gelu']
+
+_gelu_ = generate_layer_fn('gelu')
+
+
+def gelu(x):
+    locals_var = locals().copy()
+    kwargs = dict()
+    for name, val in locals_var.items():
+        if val is not None:
+            kwargs[name] = val
+    return _gelu_(**kwargs)
+
+
+gelu.__doc__ = """
+:strong:`GeLU Activation Operator`
+For more details, see [Gaussian Error Linear Units](https://arxiv.org/abs/1606.08415).
+
+Equation:
+    ..  math::
+        out = 0.5 * x * (1 + erf(\\frac{x}{\\sqrt{2}}))
+
+Args:
+
+    x(Variable): The input of GeLU op, Tensor or LoDTensor, dtype: float32 or float64.
+
+Returns:
+
+    Variable: The output of GeLU op, Tensor or LoDTensor, dtype: float32 or float64, the same as the input, shape: the same as the input.
+
+Examples:
+    
+    .. code-block:: python
+    
+        # declarative mode
+        import numpy as np
+        from paddle import fluid
+        
+        x = fluid.data(name="x", shape=(-1, 3), dtype="float32")
+        y = fluid.layers.gelu(x)
+        
+        place = fluid.CPUPlace()
+        exe = fluid.Executor(place)
+        start = fluid.default_startup_program()
+        main = fluid.default_main_program()
+        
+        data = np.random.randn(2, 3).astype("float32")
+        exe.run(start)
+        
+        y_np, = exe.run(main, feed={"x": data}, fetch_list=[y])
+        
+        data
+        # array([[ 0.87165993, -1.0541513 , -0.37214822],
+        #         [ 0.15647964,  0.32496083,  0.33045998]], dtype=float32)
+        y_np
+        # array([[ 0.70456535, -0.15380788, -0.13207214],
+        #        [ 0.08796856,  0.20387867,  0.2080159 ]], dtype=float32)
+
+    .. code-block:: python
+    
+        # imperative mode
+        import numpy as np
+        from paddle import fluid
+        import paddle.fluid.dygraph as dg
+        
+        data = np.random.randn(2, 3).astype("float32")
+        place = fluid.CPUPlace()
+        with dg.guard(place) as g:
+            x = dg.to_variable(data)
+            y = fluid.layers.gelu(x)
+            y_np = y.numpy()
+        data
+        # array([[ 0.87165993, -1.0541513 , -0.37214822],
+        #        [ 0.15647964,  0.32496083,  0.33045998]], dtype=float32)
+        y_np
+        # array([[ 0.70456535, -0.15380788, -0.13207214],
+        #        [ 0.08796856,  0.20387867,  0.2080159 ]], dtype=float32)
+"""
diff --git a/python/paddle/fluid/tests/unittests/test_gelu_op.py b/python/paddle/fluid/tests/unittests/test_gelu_op.py
new file mode 100644
index 0000000000000000000000000000000000000000..5f722ab8e03afd5768a616b6edd54247c2337f2a
--- /dev/null
+++ b/python/paddle/fluid/tests/unittests/test_gelu_op.py
@@ -0,0 +1,54 @@
+#   Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import print_function
+
+import unittest
+import numpy as np
+from scipy.special import erf
+import paddle.fluid as fluid
+import paddle.fluid.dygraph as dg
+
+
+class TestGeluOp(unittest.TestCase):
+    def _test_case1_cpu(self):
+        x = np.random.uniform(-1, 1, size=(11, 17)).astype(np.float32)
+        y_ref = 0.5 * x * (1 + erf(x / np.sqrt(2)))
+
+        place = fluid.CPUPlace()
+        with dg.guard(place) as g:
+            x_var = dg.to_variable(x)
+            y_var = fluid.layers.gelu(x_var)
+            y_test = y_var.numpy()
+        self.assertTrue(np.allclose(y_ref, y_test, rtol=1e-05, atol=1e-08))
+
+    def _test_case1_gpu(self):
+        x = np.random.uniform(-1, 1, size=(11, 17)).astype(np.float32)
+        y_ref = 0.5 * x * (1 + erf(x / np.sqrt(2)))
+
+        place = fluid.CUDAPlace(0)
+        with dg.guard(place) as g:
+            x_var = dg.to_variable(x)
+            y_var = fluid.layers.gelu(x_var)
+            y_test = y_var.numpy()
+        self.assertTrue(np.allclose(y_ref, y_test, rtol=1e-05, atol=1e-08))
+
+    def test_cases(self):
+        self._test_case1_cpu()
+        if fluid.is_compiled_with_cuda():
+            self._test_case1_gpu()
+
+
+if __name__ == '__main__':
+    unittest.main()