diff --git a/paddle/fluid/operators/conv_transpose_cudnn_op.cu b/paddle/fluid/operators/conv_transpose_cudnn_op.cu
index 15e8f3831232762eeb09bff569800cfba6cddd27..d6c7f964c68e8f80d79064dc7f6dab7767a601be 100644
--- a/paddle/fluid/operators/conv_transpose_cudnn_op.cu
+++ b/paddle/fluid/operators/conv_transpose_cudnn_op.cu
@@ -72,7 +72,7 @@ class CUDNNConvTransposeOpKernel : public framework::OpKernel<T> {
     const T* filter_data = filter->data<T>();
     const std::string data_layout_str = ctx.Attr<std::string>("data_format");
     const paddle::operators::DataLayout data_layout =
-        (data_layout_str == "NCHW" ? DataLayout::kNCHW : DataLayout::kNHWC);
+        (data_layout_str != "NHWC" ? DataLayout::kNCHW : DataLayout::kNHWC);
 
     // if channel_last, transpose to channel_first
     Tensor input_transpose;