diff --git a/src/operator/nn/lrn.cc b/src/operator/nn/lrn.cc
index 6b3d7c818378..28da2712e373 100644
--- a/src/operator/nn/lrn.cc
+++ b/src/operator/nn/lrn.cc
@@ -204,6 +204,8 @@ NNVM_REGISTER_OP(_backward_LRN)
 .set_attr<nnvm::TIsBackward>("TIsBackward", true)
 #if MXNET_USE_MKLDNN == 1
 .set_attr<FComputeEx>("FComputeEx<cpu>", LRNGradComputeExCPU)
+// Native compute requires norm while MKLDNN does not so cannot be compared in debug mode
+.set_attr<nnvm::TIsBackward>("TExcludeMKLDNNDebug", true)
 #endif
 .set_attr<FCompute>("FCompute<cpu>", LRNGradCompute<cpu>);
 
diff --git a/src/operator/nn/mkldnn/mkldnn_base.cc b/src/operator/nn/mkldnn/mkldnn_base.cc
index 8ad42a46c1a0..f07e84cc1adc 100644
--- a/src/operator/nn/mkldnn/mkldnn_base.cc
+++ b/src/operator/nn/mkldnn/mkldnn_base.cc
@@ -496,6 +496,11 @@ void OpCheck::Run(mxnet::FCompute fn, const nnvm::NodeAttrs &attrs,
                   const std::vector<mxnet::NDArray> &inputs_,
                   const std::vector<mxnet::OpReqType> &req,
                   const std::vector<mxnet::NDArray> &outputs_) {
+
+  static auto& is_excluded = Op::GetAttr<nnvm::TIsBackward>("TExcludeMKLDNNDebug");
+
+  if (is_excluded.get(attrs.op, false)) return;
+
   std::vector<mxnet::TBlob> in_blobs(inputs.size());
   for (size_t i = 0; i < in_blobs.size(); i++) in_blobs[i] = inputs[i].data();
   std::vector<mxnet::TBlob> out_blobs(outputs.size());