Aims Facilitation is a key process in vegetation dynamics, driving the response to natural and anthropogenic pressures. In harsh-grazed systems, palatable plants mainly survive when nested under unpalatable tussocks and shrubs. The magnitude and direction of positive interactions are driven by resource availability, extent of herbivory and type of nurse species. We hypothesized that different combinations of disturbance and environmental stress affect community composition in the dry Puna (southern Peruvian Andes) by modifying nurse types and plant interactions in magnitude and specific associations. We investigated whether different combinations of stress and disturbance influence species richness, type and frequency of occurrence of nurse and beneficiary species and magnitude and patterns of plant interactions; whether nurse species influence these interactions and target species change their interactions under different combinations of stress and disturbance and whether plant functional traits differ in the studied communities and influence the pattern of spatial interactions. Methods We selected three plant communities subject to different precipitation and management regimes: in each we laid a number of transects proportional to its extension. Data collected include species presence/absence, type of spatial interactions with nurse species and functional traits. We calculated species richness and rarefaction patterns, described the patterns of plant–plant spatial interactions and investigated the associations between nurse and other species in the three communities using indicator species analysis (ISA). We performed ISA and correlation analysis to investigate whether plant functional traits influenced facilitative interactions. Important Findings We found that different combinations of stress and disturbance shaped a complex set of responses, including changes in the nurse species set. Nurse composition influenced magnitude and direction of plant interactions under different stress intensities. Heavy disturbance increased the relative importance of facilitation, even if the overall number of facilitated species decreased. Under equivalent disturbance regimes, increased abiotic stress led to a greater importance of facilitation. Different combinations of stress and disturbance affected the community assemblage also by changing the behaviour of some non-nurse species. Both heavy disturbance and strong stress led to a decrease of trait states; with certain combinations of stress and disturbance, preferential distribution of these states was observed. We also found that plant traits were of key importance in determining facilitative interactions. Some traits were mainly associated with one type of spatial interaction: plant architecture, life cycle and root type influenced the type of interaction between nurses and beneficiaries under different combinations of stress and disturbance. Our results also demonstrate that in plant interaction research the object of observations (species per se , species percentage, etc.) might influence outputs, and to effectively assess the impact of different stress and disturbance intensities on plant interactions it is necessary to work at the community level to consider the whole species pool.