Commonly used methods of tolerance generation for optical systems are reviewed briefly; the most well known being that of inverse sensitivity. Another method is then discussed which will minimise a 'cost' function, subject to multiple performance constaints. The optimisation will allow either the worst case values of the effects or the RSS values to fulfill the specifications. Examples of such a set will be given, and their effects on the variation of MTF illustrated. In addition, practical upper and lower tolerance limits may be imposed.