Safety efforts are traditionally determined by the results of accident and incident analysis. Such analyses point to certain human errors as the dominant source of accidents. Support systems are then developed specifically to decrease these human errors. Based on an external safety viewpoint, this applies to transportation safety as well as plant safety. What is suggested in this paper, on the basis of findings from aeronautical field studies, is a more operator-centred safety viewpoint (internal viewpoint) which could explain some current disappointments in aiding operators. Pilots take care of incidents, planning and organizing their work as a result of their metaknowledge. They know they have limited resources and try to manage these resources as well as possible in order to do a good job with minimum risk. Most pilot expertise lies in avoiding situations where they have no solution or no chance of applying known solutions. In brief, pilots know that they are rather weak in reacting to events and compensate for these weaknesses by anticipating. The key-words of such a performance model are schematics of mental representation, flexibility in planning, anticipation, and balance between risk and efficiency. Consequences for the design of assistance systems are three-fold. First is the case of assistance systems which aim at providing the pilot with more information and more intelligence. According to the human performance model, the results of such assistance are often paradoxical, improving performance but not safety. Pilots tend spontaneously to maintain the same level of (acceptable) risk to the detriment of safety improvement. Second are support systems which aim at reacting to events. Again, the result is often disappointing. The picture is that such systems are hard for humans to understand because of the different nature of reasoning. Third are assistance systems which force the pilot to follow procedures or to be passive (the case of automation). Due to the rigidity of these procedures or automations, pilots cannot apply the model of anticipation-accommodation which is the basis of their own safety strategies. Therefore, they tend to abandon their role in order to ensure safety. The conclusion suggests that a design of support systems exclusively based on an external safety viewpoint can interact with natural and ecological ways of ensuring safety. The design of future support systems must take this perspective into account better in order to respect and take benefit from natural human abilities in ensuring safety. Some possible solutions to improve the design of assistance systems are indicated in the final section.