Native rangelands of the southwest part of the province of Santiago del Estero, Argentina, are a key source of forage for cow–calf operations. The objectives of this study were to delineate the ecosystem units of the area, to describe the associated plant communities and to interpret the role that physical factors and disturbances such as fire and grazing have had in the changes of the structure of these plant communities. This information is needed for developing recommendations for grazing management, for prescribing appropriate improvement practices (e.g. shrub control, prescribed fire) and as guidelines for future research. The ecosystem was divided into smaller units using a hierarchical method, the categories of practical importance being ‘range unit’ and ‘range site’. They represent the catchment and hillslope scale of the water runoff–runon phenomenon, respectively. Vegetation was sampled using a block and cluster sampling design, registering tree, shrub, forb and grass species frequency, and the standing aerial biomass of the herbaceous layer in a sampling unit=1 ha. Environmental data (topographic position, fire frequency, current and past use, and tree and shrub cover) were also registered for each sampling unit. Indirect ordination of sampling units classified according to range units and range sites, and correlation with environmental variables were performed using multidimensional scaling (MDS) as well as the vector fitting technique. Standing forage and stocking rate were estimated from biomass data. Results indicate that ‘range site’ is the ecosystem unit that should be considered for management purposes since it correlates well with plant communities: tall, hardwood forests are located on upland sites, woodlands are located on midland sites and savannas are located on lowland sites. Dense shrub thickets dominate in areas rated in poor condition, irrespective of range site. Disturbances such as fire and current and past use have a significant positive and negative correlation with range condition, respectively, suggesting that a state and transition model would explain vegetation dynamics better than the succession model. The estimated stocking rate in lowland sites in good condition was 2 ha UG −1, while in upland sites in poor condition the stocking rate was 90 ha UG −1. Active (fire, mechanical treatments) rather than passive (grazing management) methods should be used for range improvement in order to achieve the full potential of the ecosystem.