The US FDA created the accelerated drug approval program in 1992 to give HIV-AIDS patients earlier access to potentially life-saving, though unproven, drugs. Regulators in Europe, Japan, and elsewhere created similar conditional approval programs. They can approve a drug on weak evidence and ask the drug maker to provide more evidence later to stay on the market. A regulator typically evaluates a drug for conditional approval based on scientific evidence and the benefits of early access for patients. Our model demonstrates four ways in which a regulator should optimize these programs. First, a regulator should consider development costs, because early approval provides early revenue for the drug maker which could encourage investment in products that might otherwise be neglected. In this way, the regulator buys an option on a drug that might later prove valuable. Second, our mechanism design approach shows that the regulator’s approval threshold should be low when development costs are high and opaque to the regulator. Third, approval for a limited set of patients might be sufficient to motivate investment. Fourth, a regulator should sometimes conditionally approve drugs with a negative expected value, despite objections from FDA advisors who have resigned in protest over approvals on weak evidence.