Due to the popularity of group activities in social media, group recommendation becomes increasingly significant. It aims to pursue a list of preferred items for a target group. Recently, some methods perform graph neural network on the interaction graph. However, these works ignore that the interaction relationships among groups, users, and items are heterogeneous, i.e., two objects can be connected via different paths. In this article, we propose a heterogeneous graph attention network for group recommendation. It first employs meta-path-based random walk with restart to search for strongly correlated neighbors for each node. Then, it performs a dual-hierarchical attention network to extract semantics existing in each meta-path and fuse them to obtain hybrid representation of groups and items. Extensive experiments on three public datasets demonstrate its superiority over the state-of-the-art methods for group recommendation.