The theory and practice of digital communication during the past 50 years has been strongly influenced by Shannon's separation theorem. While it is conceptually and practically appealing to separate source from channel coding, either step requires infinite delay in general for optimal performance. On the other extreme is uncoded transmission, which has no delay but is suboptimal in general. In this paper, necessary and sufficient conditions for the optimality of uncoded transmission are shown. These conditions allow the construction of arbitrary examples of optimal uncoded transmission (beyond the well-known Gaussian example).