diff --git a/torch_npu/csrc/aten/ops/normalization/BatchNormStatsKernelNpu.cpp b/torch_npu/csrc/aten/ops/normalization/BatchNormStatsKernelNpu.cpp
index 6898ea6a7d1b3c749f21092297a1de41144b7e63..f4b22607e9f5fe7b97c1cf74612ce2d7d28cd939 100644
--- a/torch_npu/csrc/aten/ops/normalization/BatchNormStatsKernelNpu.cpp
+++ b/torch_npu/csrc/aten/ops/normalization/BatchNormStatsKernelNpu.cpp
@@ -36,7 +36,7 @@ std::tuple<at::Tensor&, at::Tensor&> batch_norm_stats_out_npu_nocheck(
   }
   at::Tensor selfCp = self;
   if (self.scalar_type() != at::kFloat){
-    selfCp = NPUNativeFunctions::npu_dtype_cast(selfCp, at::kFloat);
+    selfCp = NPUNativeFunctions::npu_dtype_cast(self, at::kFloat);
   }
   OpCommand cmd1;
   cmd1.Name("ReduceMean")