There is growing interest in the effect on candidate performance of characteristics of the interlocutor in tests of speaking. A range of variables associated with the interlocutor may cause systematic variation in linguistic performance, and consequently in scores awarded. This paper hypothesises that there may be effects on performance attributable to an interaction of these variables; in the present study, the task topic, the sex of the person presenting the topic and the sex of the candidate. This investigated in the context of a tape-mediated test of speaking, where no interlocutor is actually present; instead, stimulus material is presented by one or more speakers, one of whom acts as audience for the candidate's speech. In tests of this kind it is important to consider the impact of sex of participants in 'interpersonal' tasks (such as giving advice, making suggestions, disagreeing) where candidates have to construct their discourse to suit the 'audience'. 894 students took the tape-mediated speaking component of a test designed for students graduating from Hong Kong universities. In an advice-giving task, the last of a series involving different situations and audiences, topics considered stereotypically 'male' or 'female' oriented are presented with systematic changes in the roles taken by the male and female speakers. A multi-faceted Rasch analysis examined the interaction of test taker sex, task topic, and sex of presenter/audience. The results show that there were small effects for some, but not all, of they hypothesised interactions. Examples of differential performance by male and female candidates on other tasks are also presented and discussed. The implication for task design and test content are discussed.