Conversational humour is a complex phenomenon for a number of reasons. It is ambivalent (both aggressive and benevolent), and it is contextualized and based on shared knowledge (hence difficult to grasp fully by someone outside the group). This study explores another reason for this complexity, and that is humour as a polyphonic phenomenon, a heterogeneous discourse produced, of course, by the speaker her/himself but at the same time by many other voices (Ducrot 1984) which cross and converge with the speaker’s discourse. This paper will show that what makes conversational humour more complex is not only the fact that it is a polyphonic phenomenon but, rather, that the speaker plays hide-and-seek with the various voices s/he invokes.