In the area of building simulation, occupant modeling is one of the challenging problems due to its high uncertainty as well as its significant impact on energy simulation. To handle this, a new modeling approach, so-called an agent-based simulation approach, is demonstrated in this paper. An agent-based model deals with cognitive agents endowed with a set of behaviors that are designed to mimic behaviour of humans under certain circumstances. The agent-based model also includes stochastic movement of occupants inside multiple zones. In the paper, a series of agent-based building energy simulation runs were performed and compared with traditional building simulation runs. It was found that the results of the agent-based simulation is significantly different from those of the traditional simulation runs. The proposed model may pave a new way for the next generation of building simulation.