Compared to words that are new to a discourse, repeated words are produced with reduced acoustic prominence. Although these effects are often attributed to priming in the production system, the locus of the effect within the production system remains unresolved because, in natural speech, repetition often involves repetition of referents and lexical items simultaneously. Therefore, repetition reduction could be due to repeated mention of a referent or to repetition of a word or referring expression. In our study, we use an event description task to test whether repetition reduction is due to repetition of lexical items or to repeated mention of referents. The results show that repeated lexical items lead to reduced duration and intensity even in the absence of referent repetition, whereas repeated referents lead to reduced intensity alone. The general pattern suggests that repetition reduction is due most strongly to repetition of the lexical item, rather than repeated mention of the referent.