The cold start and data sparsity problems of the recommendation system can be effectively alleviated by cross-domain recommendation. It is possible to achieve good results by combining information from several knowledge-rich domain sources to recommend users with less information in the target domain . In other cross-domain recommendations, user and domain item information are embedded separately, and messages are transmitted through information sharing or potential feature mapping. A similarity relationship between items between domains and a relationship between users' preferences for items between domains are not considered in the transmission process. This paper proposes Cross-domain recommendation of overlapping users based on self-attention graph convolution network(OUAG). We use the graph convolution neural network to extract characteristics of embedded users and items, the attention mechanism dynamically assigns weights, and captures the higher-order user preferences in the graph through the propagation layer. A large number of experiments on two real-world data sets show that our model performs better than the baseline algorithm.