A brief review is presented of recent advances in theoretical understanding of liquid water. These advances include detailed quantitative specification of water-molecule interaction potentials, and development of several techniques in statistical mechanics for predicting molecular order in the liquid. Specifically, computer simulation now plays a particularly important rôle. A labile, defective, random network picture for the liquid is the only consistently supported possibility. Implications of this picture for an explanation of hydrophobic solvation and bonding are mentioned. Furthermore, the apparent cooperative restructuring of the random network near the supercolling limit suggests that a dielectric anomaly might exist in this régime. Several areas for future development of water theory are suggested, bearing specifically on phenomena of biological interest.