There is good evidence that cotransporters of the symport type behave as molecular water pumps, in which a water flux is coupled to the substrate fluxes. The free energy stored in the substrate gradients is utilized, by a mechanism within the protein, for the transport of water. Accordingly, the water flux is secondary active and can proceed uphill against the water chemical potential difference. The effect has been recognized in all symports studied so far (Table 1). It has been studied in details for the K+/Cl- cotransporter in the choroid plexus epithelium, the H+/lactate cotransporter in the retinal pigment epithelium, the intestinal Na+/glucose cotransporter (SGLT1) and the renal Na+/dicarboxylate cotransporter both expressed in Xenopus oocytes. The generality of the phenomenon among symports with widely different primary structures suggests that the property of molecular water pumps derives from a pattern of conformational changes common for this type of membrane proteins. Most of the data on molecular water pumps are derived from fluxes initiated by rapid changes in the composition of the external solution. There was no experimental evidence for unstirred layers in such experiments, in accordance with theoretical evaluations. Even the experimental introduction of unstirred layers did not lead to any measurable water fluxes. The majority of the experimental data supports a molecular model where water is cotransported: A well defined number of water molecules act as a substrate on equal footing with the non-aqueous substrates. The ratio of any two of the fluxes is constant, given by the properties of the protein, and is independent of the driving forces or other external parameters. The detailed mechanism behind the molecular water pumps is as yet unknown. It is, however, possible to combine well established phenomena for enzymes into a working model. For example, uptake and release of water is associated with conformational changes during enzymatic action; a specific sequence of allosteric conformations in a membrane bound enzyme would give rise to vectorial transport of water across the membrane. In addition to their recognized functions, cotransporters have the additional property of water channels. Compared to aquaporins, the unitary water permeability is about two orders of magnitude lower. It is suggested that the water permeability is determined from chemical associations between the water molecule and sites within the pore, probably in the form of hydrogen-bonds. The existence of a passive water permeability suggests an alternative model for the molecular water pump: The water flux couples to the flux of non-aqueous substrates in a hyperosmolar compartment within the protein. Molecular water pumps allow cellular water homeostasis to be viewed as a balance between pumps and leaks. This enables cells to maintain their intracellular osmolarity despite external variations. Molecular water pumps could be relevant for a wide range of physiological functions, from volume regulation in contractile vacuoles in amoeba to phloem transport in plants (Zeuthen 1992, 1996). They could be important building blocks in a general model for vectorial water transport across epithelia. A simplified model of a leaky epithelium incorporating K+/Cl-/H2O and Na+/glucose/H2O cotransport in combination with channels and primary active transport gives good quantitative predictions of several properties. In particular of how epithelial cell layers can transport water uphill.