Using software for model-based usability evaluation is uncommon today, as the modelling process is considered as overhead to the actual design work. The aim of the present paper is to describe a concept, which may make modelbased usability evaluation more worthwhile and feasible. The concept is based on sampling user models based on demographic characteristics; these characteristics may help to estimate the severity of usability problems found with the help of the user model. To sample representative user models, a simple Bayesian network (BN) was constructed, holding information about age and gender distributions, and attitudes towards technology. The results of the simulation suggest that a BN is an appropriate tool to store user information for modelling purposes, and thus may improve model-based usability evaluation.