Abstract Accurate predictions of human-building interactions are essential to decrease energy waste and improve indoor comfort. Although users’ dynamic behaviour has been recognised, most of simulation programs still work using deterministic rules. This paper investigates the effectiveness of both deterministic and stochastic models in reproducing window openings, using a Post-Occupancy evaluation (POE). A summer monitoring campaign in offices settled in Mediterranean climate was performed to acquire experimental data. Results highlight that deterministic approaches poorly reflect real behaviours. Probabilistic models seem to represent effective users’ actions. Findings demonstrate the urgent need of behavioural approaches in simulations.