Influence maximization aims at finding a set of influential individuals (i.e. users, blog etc.) in a social network. Most of the existing work focused on the influence of individuals under the hypothesis that the influence relationship between the individuals is known in advance. Nonetheless, it is often the case that groups (i.e. area, crowd etc.) are only natural targets of initial convincing attempts in many real-world scenarios, such as billboards, television marketing and plague prevention. In this paper, the problem of locating the most influential groups in a network is addressed. (1) Based on the discovery of the group associations, GIC (group independent cascade) model is proposed to simulate the influence propagation process at the group granularity. (2) A greedy algorithm called CGIM (cascade group influence maximization) is introduced to determine the top-k influential groups under GIC model. Experimental results on both synthetic and real datasets verify the effectiveness and efficiency of the presented method.