This study investigates the use of tag questions (TQs) in British English fiction dialogue by making comparisons to spoken conversation. Data has been retrieved from two subcorpora of the British National Corpus (BNC): a Fiction Subcorpus and the demographic part of the spoken component. More than 2,500 TQs have been analysed for their formal features and more than 600 TQs also for their pragmatic functions. The results show that declarative tag questions (DecTQs) are underrepresented in fiction dialogue, whereas imperative tag questions (ImpTQs) are overrepresented. Moreover, several differences between the formal features and pragmatic functions of TQs in fiction dialogue and spoken conversation have been reported. In fiction, reporting clauses and comments in the narrative provide the reader with information the author believes the reader needs to interpret the dialogue in the way the author has intended; hence, fiction dialogue is enriched with information which is useful in the analysis of a linguistic phenomenon such as the TQ. For the functional analysis of TQs, a hierarchical model has been developed and applied. Most DecTQs turn out to be used rhetorically; only a minority are response-eliciting and, in fiction dialogue, a small number also exchange goods and services. The functional patterns for DecTQs are quite different in the two subcorpora. Most rhetorical DecTQs are addressee-oriented in fiction dialogue, but speaker-centred in spoken conversation. Among the response-eliciting DecTQs, there are similar proportions of confirmation-seeking DecTQs, but, in fiction dialogue, there are proportionately more confirmation-demanding DecTQs, and also a few conversation-initiating DecTQs.