Abstract: In order to solve the problem of reducing the precision of node identification between layers and within layers due to the hierarchical classification of complex networks by the classical K-shell decomposition algorithm, the network cross-layer adjacency entropy algorithm is proposed to solve this problem. Firstly, the decomposition process of K-shell decomposition algorithm is improved, and cross-layer centrality and cross-layer center degree of network are proposed to refine the importance of network node location. Secondly, the node location information and neighbor information contained in the network cross-layer centrality, neighborhood centrality and information entropy are analyzed comprehensively, and the network cross-layer adjacency entropy algorithm is proposed to identify the importance of network nodes. Finally, five kinds of networks with different topologies are compared with other algorithms in terms of monotonicity, accuracy and time performance. The experimental results show that the monotonicity of the cross-layer adjacency entropy algorithm is up to 0.9999, and the accuracy is up to 21% higher than other algorithms, which indicates that the proposed algorithm has better ability to identify network nodes.