Abstract: To improve the running reliability of cluster tools for fabrication wafers, and to reduce maintenance cost efficiently, a Markov chainbased modeling method for preventive maintenance was proposed. According to the characteristics of the cluster tools, five states are defined, such as the running state, preventive maintenance (PM) state, PM unrestore state, preventive repair (PR) state and PR unrestore state. A state transition model was built with the five states. On the basis of the proposed model, an expected preventive maintenance cost model was presented. Finally, the built model was tested with simulative experiment data. The results indicate that the Markov chainbased preventive maintenance model is valid for analyzing the expected maintenance cost rate per unit time according to different combinations of modeling parameters.