In this paper, we will present a simple yet effective way for directed Graph (digraph) Convolutional Neural Networks based on the classic Singular Value Decomposition (SVD), named SVD-GCN for digraphs. Through empirical experiments on node classification datasets, we have found that SVD-GCN has remarkable improvements in a number of graph node learning tasks and outperforms GCN and many other state-of-the-art graph neural networks.Much of the recent literature on GNN pays particular attention to explore directed graphs, which are one major type of graph-structured data [13,12].Preprint. Under review.