Evacuation models generally tend to make over-simplifications in the representation of human behaviour in fire due to the difficulties to capture its complexity. This work shows how ‘new’ discrete choice model approaches allow overcoming this issue. A description of the existing state-of-the-art of discrete choice models is introduced to highlight the potential of those models. Three case studies are introduced describing an application of the Random Utility Theory to simulate evacuees’ decisions pertaining to the strategic, tactical and operational levels. This work highlights the need for new datasets including indicators which allows estimating latent factors which may affect evacuees’ decision.