This paper introduces nucleation theory applied to crystallizing protein solutions. It is shown that the classical approach explains the available nucleation data under most conditions used for growing protein crystals for structural studies and for industrial crystallization. However, it fails to explain most experimental data on the structure of the critical clusters. It is also shown that for open systems working out of equilibrium, such as hanging-drop and counterdiffusion techniques, the geometry of the Ostwald-Myers protein solubility diagram and the number, size, and quality of the forming crystals depend not only on supersaturation but also on the rate of development of supersaturation.