Evacuation models generally include the use of distributions or probabilistic variables to simulate the variability of possible human behaviours. A single model setup of the same evacuation scenario may therefore produce a distribution of different occupant-evacuation time curves in the case of the use of a random sampling method. This creates an additional component of uncertainty caused by the impact of the number of simulated runs of the same scenario on evacuation model predictions, here named behavioural uncertainty. To date there is no universally accepted quantitative method to evaluate behavioural uncertainty and the selection of the number of runs is left to a qualitative judgement of the model user. A simple quantitative method using convergence criteria based on functional analysis is presented to address this issue. The method permits (1) the analysis of the variability of model predictions in relation to the number of runs of the same evacuation scenario, i.e. the study of behavioural uncertainty and (2) the identification of the optimal number of runs of the same scenario in relation to pre-defined acceptance criteria.