U.S. defense policy encourages the use of dual sourcing to reduce government procurement costs, but recent theoretical work raises doubts about the benefits of this policy. I study the determinants of dual sourcing and its effects on government procurement costs using a panel dataset of tactical missiles. I find dual sourcing is not driven by failures to reduce costs; instead, it is used more often after incumbent suppliers demonstrate quality control problems, and in settings where tacit collusion is likely to be difficult. After correcting for selection bias, dual sourcing is found to reduce government procurement costs significantly.