This prospective study on a medium-fidelity simulator (SimMan, Laerdal Medical Corporation, Wappingers Falls, NY, USA) examined the management of unanticipated difficult airway by 21 anaesthetists and the effect of training in this context. There were two scenarios investigated: 'cannot intubate, can ventilate' (CI) and 'cannot intubate, cannot ventilate' (CICV). Following initial evaluation, volunteers underwent training in the 'Difficult Airway Society' (DAS) algorithms and associated technical skills. At 6-8 weeks and 6-8 months, performance was compared with the initial evaluation. There was a more structured approach following training (p < 0.05), which was sustained at 6-8 months, but only for the CICV scenario (p < 0.01). In CI, use of standard and intubating laryngeal mask airway increased following training (p = 0.021). This was sustained over time (p = 0.01). In both scenarios there was a reduced incidence of equipment misuse (p < 0.0005), which was sustained over time (p < 0.0001). We conclude that simulation-based training significantly improves performance for at least 6-8 weeks. Training should be repeated at intervals of 6 months or less.