Dennett has suggested that consciousness consists of an internal narrative, ie a dynamic internal representation of an agent's world and its position in it. We will briefly defend this general idea against some of the philosophical attacks which have been made on it, but suggest that it suffers from a different kind of problem. This picture of the machinery of consciousness has an important aspect missing, since a functioning internal representation would seem to require a great deal of internal 'narrative' which is in fact below the threshhold of consciousness. We give a sketch of how the notion of a 'computational self' might overcome this problem, and show how this might arise naturally from a truth-maintenance mechanism coupled to a reflexive architecture. Finally we will suggest a methodology for connecting phenomenology to computational models of thought.