Some advocates of empirical studies of software engineering appear to be claiming that empirical studies alone can tellus how we should do software development. This paper argues that what can be learned from empirical studies, while important, is very limited. Mathematical studies and empirical studies must be seen as mutually supportive ways to increase our knowledge of software development methods. The role of empirical studies should be limited to confirming that what works in theory can actually be used (and useful) in practice.