Summary 1. The occurrence of species vulnerable to habitat fragmentation is likely to depend on the size and separation of the fragments. However, the shape of the function that relates occurrence to these landscape parameters may be affected by other factors that are less easily measured, in which case relationships with size and separation in one area may predict occurrence elsewhere only poorly. 2. We explored how well the distribution of red squirrels Sciurus vulgaris in fragmented woodlands was predicted by simple logistic regression models empirically derived in other fragmented landscapes. 3. Comparisons between predictions lead us to identify thresholds in fragment size (> 10 ha) and distance to a source (< 600 m) where the probability of squirrel occupancy was at least 0·9 in all landscapes. These values may reflect squirrel minimum habitat requirements for home range and dispersal in the worst study area. 4. For fragments < 10 ha (outside shared thresholds), models developed in a landscape could predict squirrel occupancy elsewhere only in 17% of cases, as other factors such as demography or habitat quality might become relevant in very small and isolated fragments. 5. The predictive ability for small fragments also improved when the range of fragment sizes in the area of observation fell within the range of sizes in the area where the model was developed. 6. Some models gave correct between‐year predictions of squirrel distribution in the same landscape despite noticeable changes in regional squirrel population density. 7. When size and distance thresholds were met, we found that models could be used successfully elsewhere. In addition, threshold values indicate how large forest fragments should be and how they should be arranged to favour squirrel occurrence in a landscape.