Epistemic injustice is injustice to a person in her capacity as knower. According to the standard account, this injustice is defined by the use of ill-evidenced or unreliable generalizations connecting social identity with reliability as a witness. This view faces a dilemma about true and well-evidenced statistical generalizations connecting identity with, for example, expertise. Should we say that using them to judge people’s reliability in a given matter is not an injustice, or should we say that it is an injustice and violate the Principle of Total Evidence by ignoring them? I give an account of epistemic justice that identifies the wrong we commit when we fail at it, and implies that true statistical generalizations will often make us fail at it. I then explain why naked statistical evidence is weak evidence about an individual, and that the Principle of Total Evidence is not unconditional, to argue that the conflict between our ethical and epistemic obligations is not as extreme as it appeared.