Abstract: Identifying bot accounts in social networks can protect social network operators and their users from a variety of malicious activities. The existing social robot recognition methods based on network structure ignore that most of the real networks are heterogeneous networks with different node or edge properties, and cannot make full use of the topological information of different types of nodes or edge, resulting in certain limitations in the identification of social robots. Based on motif theory, a social robot recognition method based on heterogeneous motif features is proposed to extract more detailed local information for distinguishing human users from robot users. Compared with other existing methods, the proposed method has improved in ACC (Accuracy), Precision, Recall and F1, among which ACC has increased by 17.3% and Precision by 23.3%. At the same time, the experimental results show that the proposed method exhibits stronger robustness in the face of label noise compared with other methods. This identification method can identify social robots more accurately, so as to effectively prevent malicious robots from attacking social network platforms and spreading false information or committing fraud.