While most current building simulation tools consider occupants as predictable robots the true nature of human behaviour is more complex. This article describes a set of stochastic models aimed at capturing this complexity by decoupling occupant presence from occupant behaviour, then considering separately each means of occupant interaction (use of appliances, of windows, of lighting, etc.) with the building and finally modeling each of these appropriately. The model of occupant presence is unique in that it generates time series that have proven themselves to be realistic at both hourly and daily time scales. That of window opening assigns personal levels of tolerance to each occupant who thereafter reacts to indoor stimuli. The appliance model attributes devices to a zone, then reproduces the typical use of these by the occupants present, thereby generating a realistic variety in values of energy consumption and peak loads.