In this paper, we propose a novel Geographic Multicast routing Protocol (GMP) for wireless sensor networks1. The proposed protocol is fully distributed and stateless. Given a set of the destinations, the transmitting node first constructs a virtual Euclidean Steiner tree rooted at itself and including the destinations, using a novel and highly efficient reduction ratio heuristic (called rrSTR). Based on this locally computed tree and the information regarding the locations of its immediate neighbors, the transmitting node then splits the destinations into a set of groups and calculates a next hop for each of these groups. A copy of the packet and the locations of the corresponding group of destination nodes are directed towards the corresponding hop. The simulation results on NS2 show that the average per-destination hop count obtained using GMP is comparable to the existing PBM  algorithm and significantly less than obtained by using LGS . Most significantly, GMP requires 25% less hops and energy than alternative algorithms.