Occupancy patterns in building performance simulation are typically represented via fixed diversity profiles. More recently, stochastic models have been developed to generate random non-repeating occupancy profiles. In this context, an important question concerns the implications of occupancy modelling approaches for simulation results. The present contribution involves a virtual office building for which annual and peak heating and cooling demands are simulated. Thereby, both conventional and random profiles are deployed and different levels of occupants’ interaction with building systems are modelled. For the specific case considered here, the results do not show a noticeable difference between conventional and stochastic occupancy models.