Aspect developers constantly make a range of assumptions about the context in which their aspects will be deployed; ranging from assumptions about other aspects deployed to assumptions about semantic properties of the base and the joinpoints at which an aspect is woven. Although it has been acknowledged that such assumptions need to be made explicit to validate aspects in the face of evolution (both of aspects and the base) and reuse as well as to mitigate the fragile-pointcut problem, so far no study exists that identifies the types of assumptions aspect developers make. In this paper, we present a retrospective study of three medium-sized open-source AspectJ projects and assumptions identified in these. This leads to an initial classification of assumptions that can form the basis for further research into how best to support each type of assumption.