In this paper, we discuss a number of methodological problems we have encountered in identifying and analysing metaphors in a corpus of conversations about cancer. These problems relate particularly to: (i) the boundary between the literal and the metaphorical in the identification of linguistic metaphors; (ii) the precise identification of tenor and vehicle in relation to each linguistic metaphor; (iii) the extrapolation of conceptual metaphors from linguistic metaphors; and (iv) the extrapolation of conventional metaphors from patterns in the data. We begin with a discussion of the way in which metaphors are commonly analysed within the cognitive paradigm, and introduce in detail Steen’s [From linguistic to conceptual metaphor in five steps. In: Gibbs, R., Steen, G. (Eds.), Metaphor in Cognitive Linguistics. John Benjamins, Amsterdam, p. 57, 1999] proposal for an explicit and rigorous procedure for metaphor analysis. We then present a range of examples from our corpus that pose problems at different points in Steen’s procedure, and demonstrate how different decisions in the process of analysis lead to dramatically different conclusions as to the way in which cancer appears to be metaphorically constructed in our data. In light of our discussion, we propose some adjustments to Steen’s procedure and highlight areas in which further research is needed.