Colloids display intriguing transitions between gas, liquid, solid and liquid crystalline phases. Such phase transitions are ubiquitous in nature and have been studied for decades. However, the predictions of phase diagrams are not always realized; systems often become undercooled, supersaturated, or trapped in gel-like states. In many cases the end products strongly depend on the starting position in the phase diagram and discrepancies between predictions and actual observations are due to the intricacies of the dynamics of phase transitions. Colloid science aims to understand the underlying mechanisms of these transitions. Important advances have been made, for example, with new imaging techniques that allow direct observation of individual colloidal particles undergoing phase transitions, revealing some of the secrets of the complex pathways involved.F igure 1 illustrates three types of phase diagram. The ®rst (Fig. 1a) is a simple system of hard spheres. Introducing attractions results in three-phase equilibria, as in atomic systems such as argon (Fig. 1b). With shorter-range attractions the gas±liquid (or¯uid±¯uid) equilibrium becomes metastable (Fig. 1c). This is often observed in protein systems. One might assume equilibrium diagrams show the complete picture, with perhaps the initial distance from the phase boundary (indicative of the distance to equilibrium) controlling the speed with which the equilibrium phases are attained; however, this is the exception rather than the rule.Hard-sphere colloids suspended in a solvent provide an excellent illustration of the dif®culties involved in understanding the equilibrium states and the mechanisms by which systems evolve. Entropy considerations predict that these systems will form crystals if the volume fraction is increased. Above the`freezing' volume fraction, f f 0:494, it is entropically favourable if some spheres are in a crystal, but above the`melting' volume fraction, f m 0:545, all spheres should be in a crystal (Fig. 1a). However this is not always the situation found experimentally, either because the conditions the theory assumes (low polydispersity, for example) are not satis®ed, or the dynamics of the system have dictated a different structure (note that we cannot say that this structure is the equilibrium structureÐif entropy favours crystal formation then a crystal will form eventually, however the system may remain in the less-favoured state for a signi®cant amount of time). This crystallization process is often interpreted within the familiar framework of nucleation and growth. There has been renewed interest in this mechanism recently, both with simulations 1 and experimentally 2,3 : by monitoring the individual particles undergoing the transition it is possible to evaluate and improve the model.The classical theory predicts that the free energy cost, DG, of forming a nucleus of radius r is:where g is the surface free energy, r is the density of the bulk liquid, and Dm is the chemical potential difference between the bulk solid and bulk liquid...