The occupants' health, comfort, and productivity are important objectives for green building design and operation. However, occupant behavior also has "passive" impact on the building indoor environment by generating heat, CO2, and other "disturbances". This study develops an "indirect" practical data mining approach using office appliance power consumption data to learn the occupant "passive" behavior. The method is tested in a medium office building. The average percentage of correctly classified individual behavior instances is 90.29%. The average correlation coefficient between the predicted group schedule and the ground truth is 0.94. The experimental result also shows a fairly consistent group occupancy schedule, while capturing the diversified individual behavior in using office appliances. Compared to the occupancy schedule used in the Department of Energy prototype medium office building models, the learned schedule has a 36.67-50.53% lower occupancy rate for different weekdays. The heating, ventilation, and air conditioning (HVAC) energy consumption impact of this discrepancy is investigated by simulating the prototype EnergyPlus models across 17 different climate zones. The simulation result shows that the occupancy schedules' impact on the building HVAC energy consumption has large variations for the buildings under different climate conditions. (C) 2014 Elsevier B.V. All rights reserved.