If accurate prediction is the goal, and if information about the unconditional probability of the predicted event is available, a strong case can be made for using this information, i.e. for a Bayesian approach to inference. Not so rarely, the law calls for accurate prediction, e.g. if a bailing decision hinges on an estimate of recidivism risk. Yet for other questions of law, and for the law of evidence in particular, accuracy is not the exclusive goal. Substantive law determines who should bear the risk that doubt cannot be removed. These rules decide whether several individuals, or several acts for that matter, shall be treated as members of a class. Applying Bayes’ rule also implicitly treats the person or the action in question as a member of a class. If in conflict, the normative decision of substantive law overrides Bayes’ rule, and forces judges and administrators to neglect the base rate.