S-Crystallin from octopus lens has a tertiary structure similar to sigma-class glutathione transferase (GST). However, after isolation from the lenses, S-crystallin was found to aggregate more easily than sigma-GST. In vitro experiments showed that the lens S-crystallin can be polymerized and finally denatured at increasing concentration of urea or guanidinium chloride (GdmCl). In the intermediate concentrations of urea or GdmCl, the polymerized form of S-crystallin is aggregated, as manifested by the increase in light scattering and precipitation of the protein. There is a delay time for the initiation of polymerization. Both the delay time and rate of polymerization depend on the protein concentration. The native protein showed a maximum fluorescence emission spectrum at 341 nm. The GdmCl-denatured protein exhibited two fluorescence maxima at 310 nm and 358 nm, respectively, whereas the urea-denatured protein showed a fluorescence peak at 358 nm with a small peak at 310 nm. The fluorescence intensity was quenched. Monomers, dimers, trimers, and polymers of the native protein were observed by negative-stain electron microscopic analysis. The aggregated form, however, showed irregular structure. The aggregate was solubilized in high concentrations of urea or GdmCl. The redissolved denatured protein showed an identical fluorescence spectrum to the protein solution that was directly denatured with high concentrations of urea or GdmCl. The denatured protein was readily refolded to its native state by diluting with buffer solution. The fluorescence spectrum of the renatured protein solution was similar to that of the native form. The phase diagrams for the S-crystallin in urea and GdmCl were constructed. Both salt concentration and pH value of the solution affect the polymerization rate, suggesting the participation of ionic interactions in the polymerization. Comparison of the molecular models of the S-crystallin and sigma-GST suggests that an extra ion-pair between Asp-101 and Arg-14 in S-crystallin contributes to stabilizing the protomer. Furthermore, the molecular surface of S-crystallin has a protruding Lys-208 on one side and a complementary patch of aspartate residues (Asp-90, Asp-94, Asp-101, Asp-102, Asp-179, and Asp-180) on the other side. We propose a molecular model for the S-crystallin polymer in vivo, which involves side-by-side associations of Lys-208 from one protomer and the aspartate patch from another protomer that allows the formation of a polymeric structure spontaneously into a liquid crystal structure in the lens.