Abstract: Due to the frequent partitions and intermittent connections, routing becomes one of the most challenging problems in delay tolerant networks (DTNs), which can be extensively applied in many domains, such as interplanetary networks, mobile ad hoc networks and networks in remote areas, etc. Ant colony optimization (ACO), a probabilistic technique for finding an optimal path in a graph, is widely used in many other areas. The advantages of ACO can be concluded as positive feedback, distributed calculating, and intelligent optimization. To improve the adoption to dynamic topology, this paper investigates a routing algorithm based on ant colony to reduce the delivery delay by its intelligent and self-adapting optimization. In this paper, we first model message delivery in DTNs, and then propose our Ant-colony-based routing algorithm (ACR) which is used to guide message delivery, including forwarding-based and replication-based schemes. To show the performance of our algorithm, we finally evaluate ACR on public available data set Infocom Trace and RollerNet Trace against MED, SimBet, Spray & Wait and EBR. The simulation results show that forwarding-based ACR performs at least 25.8% better than other forwarding-based algorithms in delivery delay, and replication-based ACR achieves at least 22.5% shorter delivery delay.