diff --git a/paddle/fluid/imperative/layer.h b/paddle/fluid/imperative/layer.h
index ec5fb63f0d9339e8a511a59df05f515f01822597..892c864027d110a58546b93517242bed30a68d3a 100644
--- a/paddle/fluid/imperative/layer.h
+++ b/paddle/fluid/imperative/layer.h
@@ -281,16 +281,6 @@ class VarBase {
   static ThreadSafeNameSet name_set_;
 };
 
-class Layer {
- public:
-  virtual ~Layer() {}
-
-  virtual std::vector<std::shared_ptr<VarBase>> Forward(
-      const std::vector<std::shared_ptr<VarBase>>& inputs) {
-    return {};
-  }
-};
-
 std::shared_ptr<GradOpNode> CreateGradOpNode(
     const framework::OperatorBase& op, const NameVarBaseMap& ins,
     const NameVarBaseMap& outs, const framework::AttributeMap& attrs,
diff --git a/paddle/fluid/pybind/imperative.cc b/paddle/fluid/pybind/imperative.cc
index 5ff0e58d8580174299430bab428a7374a8231c06..cad0c1e70cc03c46c6c9f0f0fbd7326b2b74c0e3 100644
--- a/paddle/fluid/pybind/imperative.cc
+++ b/paddle/fluid/pybind/imperative.cc
@@ -59,18 +59,6 @@ PyTypeObject *g_varbase_pytype = nullptr;
 
 namespace py = ::pybind11;
 
-class Layer : public imperative::Layer {
- public:
-  using imperative::Layer::Layer;  // Inherit constructors
-
-  std::vector<std::shared_ptr<imperative::VarBase>> Forward(
-      const std::vector<std::shared_ptr<imperative::VarBase>> &inputs)
-      override {
-    PYBIND11_OVERLOAD(std::vector<std::shared_ptr<imperative::VarBase>>, Layer,
-                      Forward, inputs);  // NOLINT
-  }
-};
-
 template <typename T>
 static T PyObjectCast(PyObject *obj) {
   try {
@@ -2051,18 +2039,6 @@ void BindImperative(py::module *m_ptr) {
       .def_property_readonly("type", &imperative::VarBase::Type)
       .def_property_readonly("dtype", &imperative::VarBase::DataType);
 
-  // NOTE(zhiqiu): set the metaclass of Layer.
-  // See details: https://github.com/pybind/pybind11/pull/679
-  // https://github.com/pybind/pybind11/blob/028812ae7eee307dca5f8f69d467af7b92cc41c8/tests/test_methods_and_attributes.cpp#L284
-  py::class_<imperative::Layer, Layer /* <--- trampoline*/> layer(
-      m, "Layer", py::metaclass((PyObject *)&PyType_Type));  // NOLINT
-  layer.def(py::init<>())
-      .def("forward",
-           [](imperative::Layer &self,
-              const std::vector<std::shared_ptr<imperative::VarBase>> &inputs) {
-             return self.Forward(inputs);
-           });
-
   py::class_<imperative::jit::ProgramDescTracer>(m, "ProgramDescTracer", "")
       .def("create_program_desc",
            &imperative::jit::ProgramDescTracer::CreateProgramDesc)
diff --git a/python/paddle/fluid/dygraph/layers.py b/python/paddle/fluid/dygraph/layers.py
index 662e233bd40d616d66f7807b23599f4a9d89103e..11812398ba455053bbc6fbcfbc1dfc99df31448d 100644
--- a/python/paddle/fluid/dygraph/layers.py
+++ b/python/paddle/fluid/dygraph/layers.py
@@ -78,7 +78,7 @@ class HookRemoveHelper(object):
             del hooks[self._hook_id]
 
 
-class Layer(core.Layer):
+class Layer(object):
     """
     Dynamic graph Layer based on OOD, includes the parameters of the layer, the structure of the forward graph and so on.
 
@@ -976,7 +976,7 @@ class Layer(core.Layer):
                 for prefix, layer in model.named_sublayers():
                     print(prefix, layer)
         """
-        assert (isinstance(sublayer, core.Layer) or sublayer == None)
+        assert (isinstance(sublayer, Layer) or sublayer == None)
 
         self._sub_layers[name] = sublayer
         return sublayer
@@ -1143,7 +1143,7 @@ class Layer(core.Layer):
             params[name] = None
         else:
             layers = self.__dict__.get('_sub_layers', None)
-            if isinstance(value, core.Layer):
+            if isinstance(value, Layer):
                 if layers is None:
                     raise ValueError(
                         "super(YourLayer, self).__init__() should be called first"
diff --git a/python/paddle/fluid/tests/unittests/test_imperative_basic.py b/python/paddle/fluid/tests/unittests/test_imperative_basic.py
index 3a1ff82290e0f34537fbe0dbd349d2b730c0cb8a..d6835069b9d2a0750174654b5eb19b2aa226cafe 100644
--- a/python/paddle/fluid/tests/unittests/test_imperative_basic.py
+++ b/python/paddle/fluid/tests/unittests/test_imperative_basic.py
@@ -400,8 +400,6 @@ class TestImperative(unittest.TestCase):
 
     def test_layer(self):
         with fluid.dygraph.guard():
-            cl = core.Layer()
-            cl.forward([])
             l = fluid.Layer("l")
             self.assertRaises(NotImplementedError, l.forward, [])
 
diff --git a/python/paddle/framework/io.py b/python/paddle/framework/io.py
index 8b72f05f363cba88cb0147fa96eea3a5747d04c8..8367205a7e7c2a5f77196771eb722f794ac66a30 100644
--- a/python/paddle/framework/io.py
+++ b/python/paddle/framework/io.py
@@ -253,7 +253,7 @@ def _pickle_save(obj, f, protocol):
         dispatch_table_layer[layer.__class__] = reduce_Layer
         return layer
 
-    _parse_every_object(obj, lambda v: isinstance(v, core.Layer),
+    _parse_every_object(obj, lambda v: isinstance(v, fluid.Layer),
                         create_layer_dispatch_table)
 
     def add_dispatch_table():
@@ -316,7 +316,7 @@ def _is_state_dict(obj):
     if isinstance(obj, dict):
 
         def condition(obj):
-            return isinstance(obj, (core.Layer, Program, core.VarBase,
+            return isinstance(obj, (fluid.Layer, Program, core.VarBase,
                                     core.LoDTensor, core.SelectedRows))
 
         # If the value of a dict is a core.VarBase/LoDTensor or a dict 
@@ -422,7 +422,7 @@ def _parse_every_object(obj, condition_func, convert_func):
 
 def _parse_load_result(obj, return_numpy):
     def is_layer(obj):
-        return isinstance(obj, core.Layer)
+        return isinstance(obj, fluid.Layer)
 
     def parse_layer(obj):
         temp_dict = _parse_load_result(obj.__dict__, False)