SUMMARY (1) We surveyed the distribution of water voles (Arvicola terrestris) every month at thirty-nine sites on rivers in the North Yorkshire Moors National Park, for 2 years (1985 and 1986). Detailed studies continued at a small number of selected sites until 1989. Sites were sections of river bank 100-300 m long, the distance typically occupied by discrete breeding colonies of voles in this area. (2) In the main survey period (1985 and 1986) we found ten 'core sites' with breeding colonies of voles. A further fifteen sites were classified as 'peripheral' (visited by voles, but lacking breeding colonies). We found no evidence that water voles ever visited or used the remaining fourteen sites. (3) Data on bankside vegetation, bank angles, human disturbance, and grazing regimes by domestic stock were used in principal component analysis to classify habitat characteristics associated with core, peripheral and other sites. (4) Core sites were closely grouped in principal component space, and were characterized by a high percentage of grass, steep bank angles, and relatively high layering of the vegetation. These characteristics are readily interpreted in terms of water vole biology. (5) On the basis of habitat characteristics, we identified a further twelve unoccupied or peripheral sites that should have been suitable for breeding water voles. All these sites were more isolated and/or suffered much higher predation by mink (Mustela vison) than core sites. We also observed the extermination of two former core sites as mink invaded the area between 1987 and 1989. (6) We conclude that there are gaps in the distribution of the water vole because some habitats are unsuitable (approximately 45% of sites) and of sites with suitable habitat, about 30% lack water voles because they are too isolated (reducing colonization rates) and/or suffer very high levels of mink predation (increasing extinction rates). The conservation implications of these results are briefly discussed.