We aim to demonstrate the importance of defining linguistic phenomena by usingconstructed actionor CA (i.e. a stretch of discourse that represents one role or combination of roles depicting actions, utterances, thought, attitudes and/or feelings of one or more referents) as an example. The problem is that different assumptions about CA have led to some apparent contradictions about the nature and characteristics of this phenomenon. Based on observations and analyses of British Sign Language narrative data, we outline criteria and recommendations for defining and annotating CA. We show that, in carefully defining the phenomenon in question and providing criteria for its identification, applying these criteria to usage data leads to emergence of particular types of CA. We also show how identifying these types can help resolve some of the apparent contradictions in the literature.