Nucleation, the primary step in crystallization, dictates the number of crystals, the distribution of their sizes, the polymorph selection, and other crucial properties of the crystal population. We used timeresolved liquid-cell transmission electron microscopy (TEM) to perform an in situ examination of the nucleation of lysozyme crystals. Our TEM images revealed that mesoscopic clusters, which are similar to those previously assumed to consist of a dense liquid and serve as nucleation precursors, are actually amorphous solid particles (ASPs) and act only as heterogeneous nucleation sites. Crystalline phases never form inside them. We demonstrate that a crystal appears within a noncrystalline particle assembling lysozyme on an ASP or a container wall, highlighting the role of heterogeneous nucleation. These findings represent a significant departure from the existing formulation of the two-step nucleation mechanism while reaffirming the role of noncrystalline particles. The insights gained may have significant implications in areas that rely on the production of protein crystals, such as structural biology, pharmacy, and biophysics, and for the fundamental understanding of crystallization mechanisms.nucleation | protein | lysozyme | transmission electron microscopy | in situ observation C rystallization can be divided into two processes: nucleation and crystal growth. The crystal growth process has been well examined for a long time, yet the nucleation process is not understood; for example, the nucleation rate of crystals provides a textbook example of order-of-magnitude discrepancies between theoretical predictions and experimental results. Recent proposals have attributed these discrepancies to a nonclassical nucleation pathway, along which a structured crystalline embryo forms within a highly concentrated disordered precursor (1). This mechanism was first proposed for protein crystals (2, 3). Direct observations have demonstrated its applicability to organic (4), inorganic (5, 6), and colloidal (7) crystals. In proteins, clusters of protein molecules have been suggested as precursors; these clusters have mesoscopic sizes from several tens to several hundreds of nanometers and are considered to behave like liquids. It has also been suggested that the precursor is thermodynamically stable with respect to the mother liquid phase but is metastable or unstable with respect to the crystalline phase. The latter nature of the precursor differs from the stable macroscopically dense liquid formed as a result of the liquid-liquid phase separation (8). Such protein-rich mesoscopic clusters have been observed for many proteins, primarily using optical techniques, and have been tentatively identified as precursors for crystal nucleation (9-12). Several important questions concerning this mechanism remain unanswered. First, are the observed mesoscopic clusters actually liquid-like or solid-amorphous? Second, do they play an active role in crystal nucleation? In addition, finally, do the clusters serve as classical heteroge...