Water controlling systems are important components of smart cities. As a system of Internet of Things, water controlling systems increasingly rely on many heterogeneous perception sensors, transportation tools, and application platforms for providing throughout services. In such environments, dynamic intrusion identification is of crucial to meet security goals and take measures on important nodes considering resources for quality assurance limited by time and by cost. In this paper, we introduce a Bayesian network model for intrusion identification in IoT and propose an importance index to identify important nodes for further security management when intrusion occurs. An experiment is carried out to illustrate how this model works. After formalizing a Bayesian network on a water controlling system, Bayesian inference can be performed based on conditional probability tables of nodes with parents and prior probabilities without parents, which can be acquired statistically based on historical data. This model has good potential applications on Internet of things due to its great capability of coping with thousands of sensors, tools, and platforms with Bayesian inference, and ability of dynamically identifying important nodes to improve the efficiency of security management.