Abstract: In Vehicular Ad Hoc Networks (VANETs), high mobility, high-density, and random distribution of vehicles are crucial factors affecting the performance of a routing scheme. High mobility causes frequent changes in network topology, whereas high-density causes communication congestion due to channel contention. Moreover, routing decisions based on a relay vehicle may be less optimal if we do not consider the stability and predict the position of a relay vehicle in such dynamic environments. In addition, periodic beaconing may not be attractive because it creates channel contention and degrades the message’s reliability. Owing to these factors, designing an efficient routing scheme for emergency messages is a challenging task, especially in the urban environment. In this paper, we propose a reliable relay selection scheme for Emergency Message Routing in Intermittently Connected Networks (EMR-ICN). EMR-ICN is used for both sparse and dense network environments to establish a stable routing path. We use vehicle-to-infrastructure and vehicle-to-vehicle communication models to support routing in dense and sparse networks. We adopt the position prediction and some mobility metrics for relay selection to ensure a stable routing path. By employing relative positions changes among moving vehicles with respect to a time interval, EMR-ICN predicts the relative positions of neighbor vehicles to exclude unstable neighbors from the list of candidate relay vehicles. Moreover, it selects a reliable relay among the candidate relay vehicles based on distance, movement direction, and speed variation. To minimize channel contention, we adjust the beacons interval according to the estimation of link duration between a vehicle and its neighbors. Simulation results show that EMR-ICN outperforms Greedy Perimeter Stateless Routing (GPSR) and Maxduration-Minangle-GPSR, in terms of message delivery ratio, latency, and the number of hop count.