Occupants’ use of windows can influence the building energy demand, thermal conditions and indoor air quality. Researchers have made substantial efforts to develop probabilistic models to predict the window open/closed state. However, the hierarchical data structure and the heterogeneity in occupant behaviour have been generally neglected in previous modelling efforts. Multilevel modelling can provide an appropriate framework to handle this type of data structure and variability, but this method has rarely been used in the field. This study investigated room- and apartment-level variations in the effects of outdoor environmental variables on the window open state in low-energy apartment buildings in the UK using a multilevel modelling approach. The results showed that the room-level, rather than apartment-level, variation was statistically significant. Meanwhile, the room type (i.e., living room or bedroom) did not significantly affect the relationship between outdoor environmental variables and the window open state. The strength of this study is that the modelling accounted for the hierarchical structure of the data by simultaneously considering room-and apartment- level behavioural variations. By quantifying the significant diversity of occupant behaviour in the natural ventilation of residences, future research can more accurately estimate the variation in building energy and indoor air quality impacts.