Coal mill is an essential component of a coal-fired power plant that affects the performance, reliability, and downtime of the plant. The availability of the milling system is influenced by poor controls and faults occurring inside the mills. There is a need for automated systems, which can provide early information about the condition of the mill and help operators to take informed decisions. In this paper, a model-based residual evaluation approach, which is capable of online fault detection and diagnosis of major faults occurring in the milling system, is proposed. A dynamic mathematical model of mill, which can authentically replicate the mill behavior under different conditions, is selected for residual generation. Fuzzy logic is employed for residual evaluation to determine the type and magnitude of the fault, while Bayesian network is used for troubleshooting the root cause. The proposed technique is validated using historical data of coal mills obtained from an actual coal-fired power plant in India. Two case studies are presented to demonstrate the effectiveness of the approach. The results indicate that the proposed approach has potential to provide useful information regarding the condition of the mills and can help operators to take appropriate control action timely. This application also shows that how fuzzy logic and Bayesian networks (probability theory) can complement each other and can be used appropriately to solve parts of the problem.