The increased implementation of digital systems for instrumentation and control in nuclear power plants has given rise to a heightened risk of cyber-attacks. Given the magnitude of the consequences of cyber-attacks on nuclear power plants, it is imperative that research be focused towards detecting and responding to such events. In this paper, an event classifier to differentiate between safety events and cyber-attacks in nuclear power plants is presented. Its underlying concept is to infer the state of the system by observing both physical and network behaviors during an abnormal event and to calculate the probabilities of observing such behavior in different scenarios. These probabilities are in turn used in determining the nature of the observed abnormal event i.e., cyber or safety. The Dynamic Bayesian Networks (DBNs) methodology, which is appropriate for inferring the hidden state of the system from the observed variables through probabilistic reasoning is used to perform this task.