diff --git a/caffe2/onnx/backend.cc b/caffe2/onnx/backend.cc
index 2c060f0a7d104a..fae237a49131aa 100644
--- a/caffe2/onnx/backend.cc
+++ b/caffe2/onnx/backend.cc
@@ -343,8 +343,8 @@ Caffe2Backend::get_special_operators() const {
               {"Constant", &Caffe2Backend::CreateConstant},
               {"ConstantOfShape", &Caffe2Backend::CreateConstantOfShape},
               {"Conv", &Caffe2Backend::CreateConvPoolOpBase},
-              {"AveragePool", &Caffe2Backend::CreatePadPool},
-              {"GlobalAveragePool", &Caffe2Backend::CreatePadPool},
+              {"AveragePool", &Caffe2Backend::CreateConvPoolOpBase},
+              {"GlobalAveragePool", &Caffe2Backend::CreateConvPoolOpBase},
               {"GlobalMaxPool", &Caffe2Backend::CreateConvPoolOpBase},
               {"MaxPool", &Caffe2Backend::CreateConvPoolOpBase},
               {"Reshape", &Caffe2Backend::CreateReshape},
@@ -545,65 +545,6 @@ Caffe2Ops Caffe2Backend::CreateConvPoolOpBase(
   return CommonOnnxNodeToCaffe2Ops(onnx_node, ctx);
 }
 
-Caffe2Ops Caffe2Backend::CreatePadPool(
-    OnnxNode* onnx_node,
-    const ConversionContext& ctx) {
-  auto& node = onnx_node->node;
-  auto& attributes = onnx_node->attributes;
-  Caffe2Ops ret;
-  // Pad
-  bool padding = false;
-  const std::string pad_name = ctx.opset_version() < 2 ? "paddings" : "pads";
-  const auto pad_input = dummy_->NewDummyName();
-  if (attributes.HasAttribute("count_include_pad") &&
-      attributes.HasAttribute(pad_name)) {
-    auto count_include_pad = attributes.get<int64_t>("count_include_pad", 0L);
-    ::google::protobuf::RepeatedField<::google::protobuf::int64> pads;
-    pads =
-        attributes
-            .get<::google::protobuf::RepeatedField<::google::protobuf::int64>>(
-                pad_name);
-    if (count_include_pad == 1 && pads.size() == 4 &&
-        !(pads.Get(0) == 0 && pads.Get(1) == 0 && pads.Get(2) == 0 &&
-          pads.Get(3) == 0)) {
-      padding = true;
-      attributes.remove(pad_name);
-      caffe2::Argument arg_pads;
-      arg_pads.add_ints(pads.Get(0));
-      arg_pads.add_ints(pads.Get(1));
-      arg_pads.add_ints(pads.Get(2));
-      arg_pads.add_ints(pads.Get(3));
-      arg_pads.set_name("pads");
-      auto* c2_op = ret.ops.Add();
-      BuildOperator(
-          c2_op, "PadImage", {node.input(0)}, {pad_input}, {arg_pads});
-    } else if (count_include_pad == 1) {
-      std::string str;
-      bool pads_flag = false;
-      str += "[";
-      for (const auto& i : pads) {
-        str += c10::to_string(i) + ",";
-        pads_flag = pads_flag || i > 0;
-      }
-      str += "]";
-      if (pads_flag == true) {
-        CAFFE_THROW(
-            "Caffe2 only supports padding 2D Tensor, whereas padding is ", str);
-      }
-    }
-  }
-  // Pool
-  auto c2_ops = Caffe2Backend::CreateConvPoolOpBase(onnx_node, ctx);
-  auto* pool_op = c2_ops.ops.Mutable(0);
-  if (padding) {
-    pool_op->set_input(0, pad_input);
-  }
-  auto* c2_op = ret.ops.Add();
-  c2_op->CopyFrom(*pool_op);
-
-  return ret;
-}
-
 Caffe2Ops Caffe2Backend::CreateReshape(
     OnnxNode* onnx_node,
     const ConversionContext& ctx) {
diff --git a/test/onnx/test_pytorch_onnx_caffe2.py b/test/onnx/test_pytorch_onnx_caffe2.py
index 58f255f3f36c8a..c31d7c2bf66766 100644
--- a/test/onnx/test_pytorch_onnx_caffe2.py
+++ b/test/onnx/test_pytorch_onnx_caffe2.py
@@ -664,11 +664,18 @@ def test_maxpool2d_single_padding(self):
         model = nn.MaxPool2d(5, padding=2)
         self.run_model_test(model, train=False, batch_size=BATCH_SIZE)
 
-    @unittest.skip("C2 and PyTorch have small difference in padding implementation")
     def test_avgpool2d(self):
         model = nn.AvgPool2d(5, padding=(2))
         self.run_model_test(model, train=False, batch_size=BATCH_SIZE)
 
+    def test_avgpool2d_with_count_include_pad_set_false(self):
+        model = nn.AvgPool2d(7, padding=(2), count_include_pad=False)
+        self.run_model_test(model, train=False, batch_size=BATCH_SIZE)
+
+    def test_avgpool2d_with_count_include_pad_set_true(self):
+        model = nn.AvgPool2d(7, padding=(2), count_include_pad=True)
+        self.run_model_test(model, train=False, batch_size=BATCH_SIZE)
+
     def test_avgpool2d_no_padding(self):
         model = nn.AvgPool2d(5)
         self.run_model_test(model, train=False, batch_size=BATCH_SIZE)