‘Bias’ can be used to describe a problematic kind of prejudice against a particular social group, but it can also be used more generally, and neutrally, to describe a process that is systematically weighted in some way. This chapter takes up the question of what distinguishes instances of “prejudicial” bias in the context of the visual system, a systematically, formally biased system. Is prejudicial bias marked out by a set of distinctive structural features that explain the epistemic problems intuitively associated with it, or is it distinguished only by the way in which its content concerns particular demographic categories? I take two bodies of recent empirical work as case studies, the “same race face effect”, and evidence and analysis of racial bias in first-person shooter tasks. After exploring and rejecting three possible criteria for demarcating instances of prejudicial bias, I identify a principled reason why standard epistemic criteria cannot accomplish the task. I instead propose an adoption of a skill-based model of visual perception that allows for multi-dimensional evaluation relative to a set of potentially competing goals.