Building occupancy is an important basic factor in building energy simulation but it is hard to represent due to its temporal and spatial stochastic nature. This paper presents a novel approach for building occupancy simulation based on the Markov chain. In this study, occupancy is handled as the straightforward result of occupant movement processes which occur among the spaces inside and outside a building. By using the Markov chain method to simulate this stochastic movement process, the model can generate the location for each occupant and the zone-level occupancy for the whole building. There is no explicit or implicit constraint to the number of occupants and the number of zones in the model while maintaining a simple and clear set of input parameters. From the case study of an office building, it can be seen that the model can produce realistic occupancy variations in the office building for a typical workday with key statistical properties of occupancy such as the time of morning arrival and night departure, lunch time, periods of intermediate walking-around, etc. Due to simplicity, accuracy and unrestraint, this model is sufficient and practical to simulate occupancy for building energy simulations and stochastic analysis of building heating, ventilation, and air conditioning (HVAC) systems.