Standards and methods for the assessment of surgical skills are under increased scrutiny as a result of public and professional reactions to clinical failures1. Concerns are that surgeons have a false perception of their abilities, leading to avoidable deaths. To prevent this, better surgical skills assessment has been proposed1. Comparisons between surgery and aviation as two hazardous industries have led to the recommendation of mandatory simulation training before patient encounters2, but the disciplines of aviation and surgery are not strictly comparable. In contrast to aviation simulators, surgical simulators are incomplete to equip a surgeon with sufficient skills to take directly to a real-life operating theatre. The complexity and variation in human anatomy, physiology and disease inevitably means that a fully standardized approach cannot apply to every operation. The skills needed for a non-standardized surgical environment are different from those of a standardized aviation environment. Although surgeons and pilots share a thinking process to determine the correct action by interpreting behavioural clues, the necessary coordination in aviation is explicit, resulting from preset instructions in a highly standardized environment3. Some assumptions about simulated surgical skills training and translation into clinical practice do not stand up to close examination. The autopilot has become a standard component in large aeroplanes. The first fully automated transatlantic flight took place in 1947. Autopilot is the only way to control aeroplanes in many flight phases owing to the effect of temperature on aeroplane parts and surrounding air turbulence4. Simulation is the only way to train commercial pilots in the absence of real flight training opportunities in the air. Robotic simulation is often portrayed as an analogous device for the surgeon, but there is no form of autosurgeon. Robots available for surgical procedures are simple slave mechanical devices that work only with direct human interaction. Another belief is that practice on real patients is dangerous. There is no evidence that supervised surgical training compromises patient safety. Teaching hospitals have better or comparable outcomes to nonteaching hospitals after case-mix correction5. It is 26 times safer to fly than to drive a car6, so perhaps it might be more reasonable to compare surgery with driving. It is still standard to learn driving with an instructor on the road. Drivers, like surgeons, operate in a minimally standardized environment with many variations. Drivers have an agreed thinking process to avoid hazards faced, but no step-by-step instructions. Surgeons have to deal with anatomical and pathological variations, along with unexpected events in the operating theatre, in the same way that drivers have to respond to unexpected events on the road. Driving test assessment tools provide further insight into this similarity. In many countries, the driving test examines both theoretical knowledge and practical skill7. In the UK, the theoretical part consists of multiple choice questions (MCQ) to assess knowledge and hazard perception clips to check learners’ awareness, followed by a period of instructorfacilitated training on the roadwhen it is illegal for the learner to drive alone. Once the learner is fully trained, the practical test is taken. This involves an on-road driving assessment with an examiner armed with a checklist. The clinical skills literature supports the relevance of this type of assessment for the development of surgical skills. Miller8 provided a pyramidal structure to assess clinical skills, with lower levels about knowledge covered by MCQ. Surgical training and assessment are performed in the operating theatre by the trainer using procedure-based assessment forms9. This covers the remaining practical skills assessment levels, with clear parallels to the practical driving test. In surgical training, hazard perception might be achieved with error recovery training10. This stresses subjective detection of errors committed by a third party while using content to justify own errors. Dror10 suggested the use of interactive training clips to detect errors in others, linking error detection with the search for a recovery plan. Making these links should replace self-error justification and denial mechanisms. This new cognitive process could reduce errors and mitigate damage during operations. Errors are infrequent in the standardized automated