As referents are more accessible in discourse, they can be referred to with more attenuated expressions, such as pronouns. Accessibility is known to be affected by the referents saliency in the linguistic context, but much less is known about the effect of saliency in the visual context. In this paper, we investigate whether a referents visual saliency affects the choice of referring expression in a discourse context. The results of a story completion experiment show that visually salient referents induce more attenuated expressions, but only when they are linguistically non-salient. Linguistically salient referents receive more reduced references when they are visually non-salient. We argue that visual saliency affects accessibility when the impact of linguistic factors is moderated. In addition, when the story does not match peoples expectations, processing difficulties might result in the use of less costly expressions.