Graphene oxide membranes show exceptional molecular permeation properties, with a promise for many applications. However, their use in ion sieving and desalination technologies is limited by a permeation cutoff of 9 Å, which is larger than hydrated ion diameters for common salts. The cutoff is determined by the interlayer spacing d 13.5 Å, typical for graphene oxide laminates that swell in water. Achieving smaller d for the laminates immersed in water has proved to be a challenge. Here we describe how to control d by physical confinement and achieve accurate and tuneable ion sieving. Membranes with d from 9.8 Å to 6.4 Å are demonstrated, providing the sieve size smaller than typical ions' hydrated diameters. In this regime, ion permeation is found to be thermally activated with energy barriers of 10-100 kJ/mol depending on d. Importantly, permeation rates decrease exponentially with decreasing the sieve size but water transport is weakly affected (by a factor of <2). The latter is attributed to a low barrier for water molecules entry and large slip lengths inside graphene capillaries. Building on these findings, we demonstrate a simple scalable method to obtain graphene-based membranes with limited swelling, which exhibit 97% rejection for NaCl.Selectively permeable membranes with sub-nm pores attract strong interest due to analogies with biological membranes and potential applications in water filtration, molecular separation and desalination [1][2][3][4][5][6][7][8] . Nanopores with sizes comparable to, or smaller than, the diameter D of hydrated ions are predicted to show enhanced ion selectivity 7,9-12 because of dehydration required to pass through such atomic-scale sieves. Despite extensive research on ion dehydration effects 3,7,9-13 , experimental investigation of the ion sieving controlled by dehydration has been limited because of difficulties in fabricating uniform membranes with well-defined sub-nm pores. The realisation of membranes with dehydration-assisted selectivity would be a significant step forward. So far, research into novel membranes has mostly focused on improving the water flux rather than ion selectivity. On the other hand, modelling of practically relevant filtration processes shows that an increase in water permeation rates above the rates currently achieved (2-3 L/m 2 ×h×bar) would not contribute greatly to the overall efficiency of desalination 8,14,15 . Alternative approaches based on higher water-ion selectivity may open new possibilities for improving filtration technologies, as the performance of state-of-the-art membranes is currently limited by the solution-diffusion mechanism, in which water molecules dissolve in the membrane material and then diffuses across the membrane 8 . Recently, carbon nanomaterials including carbon nanotubes (CNT)