Graph kernels have achieved excellent performance in graph classification tasks. In this paper, we propose a novel deep motif entropy graph kernel for the purpose of graph classification. For better capturing the differences between substructures, we gauge detailed information through a family of K-layer expansion motifs rooted at each node and combine the Weisfeiler-Lehman algorithm to subdivide motifs, which is further enhanced by motif entropy. Experiments on eight graph-structured datasets demonstrate that our method is able to outperform the state-of-the-art kernel methods for the tasks of graph classification.