Delay/Disruption Tolerant Network (DTN) has characteristics of long delay, intermittent disruption, and limitation of buffer space and energy. To improve the delivery rate of messages, while reducing network overhead and the average latency, a new Routing Algorithm Based on Node Similarity (RABNS) in DTN was proposed. The algorithm used historical information to predict node encounter probability in future. The nodes which encountered historically were recorded as a collection, then the set intersection operation was applied to evaluate the similarity of a pair of nodes. And the similarity was used to control the number of copies in the network. Simulations were conducted on The ONE platform using RandomWaypoint motion model. In the simulation, RABNS performed better than PROPHET (Probabilistic ROuting Protocol using History of Encounters and Transitivity) in the message delivery rate. And the network overhead of RABNS was about half of PROPHET, which greatly improved the utilization of network resources. The average latency of RABNS was a little higher than Epidemic but lower than PPROPHET, the node cache size did not have a significant impact on average-hops, and its average-hops was about half of PROPHET. The simulation results show that RABNS can effectively limit the message flooding with higher message delivery rate, lower network overhead and average latency, therefore it is suitable for the DTN scenes with limited nodes' storage and also applicable in social DTN with gregarious characteristics.