diff --git a/paddle/operators/pad_op.cc b/paddle/operators/pad_op.cc
index 99f605c651d7f48916f8735a1c4948c0d71fa613..7e78b6ec133981494a65b5e16316ae8fdbd61a60 100644
--- a/paddle/operators/pad_op.cc
+++ b/paddle/operators/pad_op.cc
@@ -96,8 +96,9 @@ class PadOpGrad : public framework::OperatorWithKernel {
                             "Input(Out@GRAD) should not be null");
     auto x_dims = ctx.Input<Tensor>("X")->dims();
     auto *x_grad = ctx.Output<Tensor>(framework::GradVarName("X"));
-
-    x_grad->Resize(x_dims);
+    if (x_grad != nullptr) {
+      x_grad->Resize(x_dims);
+    }
   }
 };
 
diff --git a/paddle/operators/pad_op.h b/paddle/operators/pad_op.h
index ca8832f26a6492bf265b07e1581a886777e6301e..2cc3b945ae5b2e2e93d8531c7f99e4c215d1d806 100644
--- a/paddle/operators/pad_op.h
+++ b/paddle/operators/pad_op.h
@@ -87,12 +87,13 @@ void PadGradFunction(const framework::ExecutionContext& context) {
   }
   auto* d_out = context.Input<Tensor>(framework::GradVarName("Out"));
   auto* d_x = context.Output<Tensor>(framework::GradVarName("X"));
-  d_x->mutable_data<T>(context.GetPlace());
-
-  auto d_x_tensor = EigenTensor<T, D>::From(*d_x);
-  auto d_out_tensor = EigenTensor<T, D>::From(*d_out);
-  auto place = context.GetEigenDevice<Place>();
-  d_x_tensor.device(place) = d_out_tensor.pad(paddings, 0);
+  if (d_x != nullptr) {
+    d_x->mutable_data<T>(context.GetPlace());
+    auto d_x_tensor = EigenTensor<T, D>::From(*d_x);
+    auto d_out_tensor = EigenTensor<T, D>::From(*d_out);
+    auto place = context.GetEigenDevice<Place>();
+    d_x_tensor.device(place) = d_out_tensor.pad(paddings, 0);
+  }
 }
 
 template <typename Place, typename T>