Building performance is influenced by occupants' presence and actions. Knowledge of occupants' future presence and behaviour in buildings is of central importance to the implementation efforts concerning predictive building systems control strategies. Specifically, prediction of occupants' presence in office buildings represents a necessary condition for predicting their interactions with building systems. In the present contribution, we focus on the evaluation of a number of occupancy models to explore the potential of monitored past occupancy data towards predicting future presence of occupants. Towards this end, we obtained long-term high-resolution monitored occupancy data from a number of workplaces (in open, semi-open, and closed office settings) in a university building. Using this data, we trained two existing probabilistic occupancy models and an original non-probabilistic occupancy model to predict the occupancy profiles of the same workplaces on a daily basis. The predictions were evaluated via comparison with monitored daily occupancy profiles. To conduct the model evaluation in a rigorous manner, separate sets of data were used to train and evaluate the models. A set of five specific evaluation statistics was deployed for model comparison. In general, the obtained level of predictive accuracy of all models considered was found to be rather low. However, the proposed non-probabilistic model performed better in view of short-term occupancy predictions. The results thus facilitate a discussion of the potential and limitations of predicting building occupants' future presence patterns based on past monitoring data. (C) 2014 Elsevier B.V. All rights reserved.