Abstract: A novel method for group activity recognition in crowd is proposed using causality analysis. The Granger Causality Test is used to analyze the causality between individual actions. On this basis, we adopt a dominant set based clustering algorithm to detect interacting groups in crowded scenes using causality, spatial and directional relationships among people. To effectively represent group activity, low level visual features and causality features are used. The low level visual features, which included histograms of oriented gradients (HOG) and histograms of optical flow (HOF), are applied to describe the properties of individual activity, and the causality features obtained by causality analysis are introduced to depict the interaction information of people. Sparse representation is employed to recognize group activities in crowd. Experiments are performed on the BEHAVE and collective activity databases to test and evaluate the proposed method. The experiments results show that the proposed method is more effective than other state-of-the-art methods.