Considering the characteristics of the interaction in the current volleyball group behavior recognition methods, such as directed, delay and subject to spatio-temporal constraints, this paper proposes a volleyball group behavior recognition method considering multi-scale spatio-temporal causal features. Firstly, the causal relationship of behavior interaction was analyzed and judged based on the causal detection model. Secondly, based on the temporal causality diagram reasoning model, the temporal causal features in crowd behavior were extracted to reduce the error caused by the delay pair feature fusion process. Then, the distance and appearance constraints are introduced into the temporal causal features to extract the multi-scale spatio-temporal causal features of crowd behavior. Finally, the effectiveness of the proposed algorithm is verified by public datasets and self-selected datasets, and the importance of the modules in the recognition framework is verified by fusion and comparison experiments. The experimental results show that the proposed model can give full play to the advantages of multi-scale spatio-temporal causal maps in describing group interaction, and the extracted multi-scale spatio-temporal causal feature maps can effectively learn the characteristics of crowd behavior. It can improve the performance of volleyball group behavior recognition.