Discrimination is the unjust treatment of people based on their group membership, and detecting discrimination is a necessary first step in countering it. Attributions to discrimination have often been conceptualized in terms of Signal Detection Theory (SDT), entailing speculation and theorizing about the mediation of observed effects on judgments through observers’ sensitivity (i.e., their ability to recognize discrimination) or response bias (i.e., their tendency to judge that it is present). Here, we take a first step toward moving beyond conceptual applications to empirical SDT analyses of attributions to discrimination. Across five studies examining judgments about gender discrimination in corporate pay raise decisions, we find that prototypes (Studies 1 to 3) and their alignment with the actual pattern of decisions (Studies 2 to 4) inform observers’ response bias, reflecting differences in their attribution of unfair outcomes to discrimination (Study 5). In contrast, the extent of harm caused by discriminatory outcomes exerts a small effects on sensitivity and a large effect on response bias (Studies 1 to 3), which reflects observers’ determination of a fair standard for decisions and not their attributions (Study 5). We generally find that response bias is more malleable than sensitivity — which is a theoretical advancement but poses challenges for training initiatives aimed at reducing discrimination through increasing awareness. Taken together, our results exemplify and highlight the complex and often distinct mechanisms through which variables such as harm or prototypes shape attributions to discrimination.