The principles of statistical mechanics have been used to develop a theory of the nucleation of a phase transition, but a number of subtle questions remain and are highlighted in this review. A central issue is the cluster definition, the mathematical scheme which distinguishes a molecular cluster from a collection of separate molecules. There is also the question of whether thermodynamic transition state theory is suitable to describe the process, or whether the development of kinetic rate equations describing cluster growth and decay is a better approach. The classical theory of nucleation is flawed but appears to provide useful estimates in some cases, including water. Phenomenological extensions of the classical theory can improve matters. However, improvements in the theory from microscopic considerations are not simple to apply, require major computational effort and suffer from uncertainties due to lack of knowledge of the fundamental intermolecular interactions. Calculations of the nucleation rate of water droplets are especially difficult since this substance is notoriously difficult to model. Nevertheless, as capabilities improve, accurate calculations should come within reach, which will offer better understanding of the process for practical applications such as the transition from dry to wet steam.