Data from many domains can be represented as graphs growing over time.Examples include collaborations among scientists, friendships on social networking websites, etc. A common task is to predict new links based on the current state of the graph. Our main contribution is using temporal information to improve the performance of existing link prediction algorithms. Given a central node and its neighborhood, we build a probabilistic model based on the principle of maximum entropy and use it to rank neighbors according to probability that they we be linked with the central node.