•  We test a novel model of osmosis in aquaporins. •  A solute molecule present at the pore mouth can be reflected or permeate the pore; we propose that only reflected molecules induce osmotic water transport, while permeating molecules give rise to no water transport. •  We tested a range of channel geometries using aquaporins AQP1 and AQP9 and mutants thereof; the aquaporins were expressed in Xenopus oocytes. Osmotic gradients were generated by solutes of molecular weights in the range 45–595 Daltons. The reflected fraction of a given solute was estimated optically and compared to the permeability obtained from uptakes of radio‐labelled solutes. •  In accordance with our model there was a linear relationship between solute permeability and reflection coefficient: solutes with high permeability had low reflection coefficients and vice versa. •  We found no evidence for significant coupling between water and solute fluxes inside the pore. Abstract  We test a novel, stochastic model of osmotic water transport in aquaporins. A solute molecule present at the pore mouth can either be reflected or permeate the pore. We assume that only reflected solute molecules induce osmotic transport of water through the pore, while permeating solute molecules give rise to no water transport. Accordingly, the rate of water transport is proportional to the reflection coefficient σ, while the solute permeability, PS, is proportional to 1 –σ. The model was tested in aquaporins heterologously expressed in Xenopus oocytes. A variety of aquaporin channel sizes and geometries were obtained with the two aquaporins AQP1 and AQP9 and mutant versions of these. Osmotic water transport was generated by adding 20 mm of a range of different‐sized osmolytes to the outer solution. The osmotic water permeability and the reflection coefficient were measured optically at high resolution and compared to the solute permeability obtained from short‐term uptake of radio‐labelled solute under isotonic conditions. For each type of aquaporin there was a linear relationship between solute permeability and reflection coefficient, in accordance with the model. We found no evidence for coupling between water and solute fluxes in the pore. In confirmation of molecular dynamic simulations, we conclude that the magnitude of the osmotic water permeability and the reflection coefficient are determined by processes at the arginine selectivity filter located at the outward‐facing end of the pore.