In conventional networks, such as the Internet, it is assumed that there exists a fully connected path between any pair of nodes at any time. In Delay Tolerant Networks (DTN), however, the connected path may be broken, and nodes may be partitioned into two disconnected subsets because of the movement or failure of some nodes. Thus, the data transmission requirements can not be satisfied by traditional routing schemes. How to efficiently route data in DTNs has therefore become a critical and challenging issue. To address the problem, we propose a contact-duration-based probabilistic routing scheme (PR_CD) based on the probabilistic routing scheme PRoPHET (PR). Each node in the PR_CD scheme predicts the current network condition by using contact information, and then chooses appropriate relay nodes to improve the delivery rate as well as reduce the transmission latency and transmission overhead. To evaluate the efficacy of the proposed scheme, we use simulations to conduct a comprehensive study of different DTN routing mechanisms. We believe such a study is important because the ability to characterize the behavior and performance of routing schemes plays a key role in the understanding and design of DTNs.