Based on almost seven years of continuous measurements we have analysed in detail the influence of occupancy patterns, indoor temperature and outdoor climate parameters (temperature, wind speed and direction, relative humidity and rainfall) on window opening and closing behaviour. This paper presents the development and testing of several modelling approaches, including logistic probability distributions, Markov chains and continuous-time random processes. Based on detailed statistical analysis and cross-validation of each variant, we propose a hybrid of these techniques which models stochastic usage behaviour in a comprehensive and efficient way. We conclude by describing an algorithm for implementing this model in dynamic building simulation tools.