In 2013, the Supreme Court made the offhand comment that empirical models and their estimations or predictions are not “findings of fact” deserving of deference on appeal. The four Justices writing in dissent disagreed, insisting that an assessment of how a model works and its ability to measure what it claims to measure are precisely the kinds of factual findings that the Court, absent clear error, cannot disturb. Neither side elaborated on the controversy or defended its position doctrinally or normatively. That the highest Court could split 5–4 on such a crucial issue without even mentioning the stakes or the terms of the debate, suggests that something is amiss in the legal understanding of models and modeling.

This Article does what that case failed to do: it tackles the issue head-on, defining the legal status of a scientific model’s results and of the assumptions and choices that go into its construction. I argue that as a normative matter, models and their conclusions should not be treated like facts. Models are better evaluated by a judge, they do not merit total deference on appeal, and modeling choices are at least somewhat susceptible to analogical reasoning between cases. But I show that as a descriptive matter, courts often treat models and their outcomes like issues of fact, despite doctrines like Daubert that encourage serious judicial engagement with modeling. I suggest that a perceived mismatch between ability and task leads judges to take the easier route of treating modeling issues as facts, and I caution that when judges avoid hard questions about modeling, they jeopardize their own power and influence. . . .