1. The stress gradient hypothesis suggests a shift from predominant competition to facilitation along gradients of increasing environmental severity. This shift is proposed to cause parallel changes from prevailing spatial segregation to aggregation among the species within a community. 2. We used 904 1-m(2) plots, each subdivided into 100 10 x 10 cm, or 25 20 x 20 cm cells, respectively, from 67 European mountain summits grouped into 18 regional altitudinal transects, to test this hypothesized correlation between fine-scale spatial patterns and environmental severity. 3. The data were analysed by first calculating standardized differences between observed and simulated random co-occurrence patterns for each plot. These standardized effect sizes were correlated to indicators of environmental severity by means of linear mixed models. In a factorial design, separate analyses were made for four different indicators of environmental severity (the mean temperature of the coldest month, the temperature sum of the growing season, the altitude above tree line, and the percentage cover of vascular plants in the whole plot), four different species groups (all species, graminoids, herbs, and all growth forms considered as pseudospecies) and at the 10 x 10 cm and 20 x 20 cm grain sizes. 4. The hypothesized trends were generally weak and could only be detected by using the mean temperature of the coldest month or the percentage cover of vascular plants as the indicator of environmental severity. The spatial arrangement of the full species set proved more responsive to changes in severity than that of herbs or graminoids. The expected trends were more pronounced at a grain size of 10 x 10 cm than at 20 x 20 cm. 5. Synthesis. In European alpine plant communities the relationships between small-scale co-occurrence patterns of vascular plants and environmental severity are weak and variable. This variation indicates that shifts in net interactions with environmental severity may differ among indicators of severity, growth forms and scales. Recognition of such variation may help to resolve some of the current debate surrounding the stress gradient hypothesis.