Perception is a safety-critical function of autonomous vehicles and machine learning (ML) plays a key role in its implementation. This position paper identifies (1) perceptual uncertainty as a performance measure used to define safety requirements and (2) its influence factors when using supervised ML. This work is a first step towards a framework for measuring and controling the effects of these factors and supplying evidence to support claims about perceptual uncertainty.