We experimentally investigate the determinants of judicial decisions in a setting resembling real-world judicial decision-making. U.S. federal judges (N=32) spend 55 minutes judging a real appeals case from an international tribunal, with minor modifications to accommodate the experimental treatments. The fictitious briefs focus on one easily understandable issue of law. Our 2×2 between-subject factorial design crosses a weak precedent and legally irrelevant defendant characteristics. In a survey, law professors predicted that the precedent would have a stronger effect than the defendant characteristics. In actuality, the precedent has no detectable effect on the judges’ decisions, whereas the two defendants’ affirmance rates differ by 45% (p<.01). Judges’ written reasons, on the other hand, do not mention defendant characteristics at all, focusing instead on the precedent and other legalistic and policy considerations.