When a speaker produces a referring expression, their overarching goal is to get the addressee to identify a particular object in the context. This goal leads to the expectation that speakers will use a referring expression tailored to the perspective of the addressee. While research in psycholinguistics has indeed found that speakers tailor their referring expressions to the addressee’s perspective, they also find egocentric tendencies; namely, a sensitivity to the speaker’s own perspective. Mozuraitis, Stevenson and Heller (2018) make the novel proposal that “mixing” perspectives is a design feature of the production system, modelling data from an experiment where knowledge mismatch concerned object function. Here we further test this model on the more common knowledge mismatch of visual perspective, modelling data from Vanlangendonck, Willems, Menenti and Hagoort (2016). The modelling results shed new light on concept of “referential success” that has been assumed to guide reference production.