Tracking the evolution and detecting events are popular and difficult problems in the field of social network analysis. Most of the research focuses on proposing different models to fit different network characteristics. This type of approach usually has three problems: (1) Each model is designed for one particular network and cannot well fit other networks; (2) There are many network statistics, so the evaluation of these network models lacks of unified platforms; (3) Without taking temporal information into account, these network models can hardly track the evolution and detect events. To solve these problems, this paper presents a method for event detection in social networks based on link prediction, which can evaluate the fluctuation of the networks and detect the events in social networks. The main work is as follow: (1) Demonstrates the method "modelling and evaluating" is in accord with link prediction on revealing the network evolution mechanism; (2) Proposes an algorithm similarity computing (SimC) to compute the similarity of networks and further improves this algorithm by taking micro factors into account; (3) Evaluates the fluctuation of the network evolution and proposes an event detecting (EventD) algorithm to detect the events. The results of the experiment show that the presented method can effectively solve the problem of tracking the evolution and detecting events.