Based on 6 years of continuous measurements, we have analysed in detail the occupancy, thermal and visual parameters influencing actions on shading devices in order to derive an accurate model for the prediction of their usage in office buildings. This article begins by presenting some of the key findings from these analyses. Informed by other developments in the literature, we go on to propose an approach for a comprehensive stochastic model for simulating blind usage. This model is based on a Markov process taking rigorously selected predictors (initial blind status, indoor and outdoor illuminance) as input variables to predict lowering and raising actions performed by occupants. A separate sub-model then predicts the chosen shaded fraction. An assessment of the predictive accuracy of simulations is then presented for several modelling variants using our measured data, from which the best performing model variant is selected.