We quantify the worst-case price of fairness for matroid allocation problems in various settings, from adversarial, to semi random, to fully random.