We investigated relationships between rainfall (and landscape, zonation and nearby grazing disturbance) and the death rates of four perennial grass species in a highly functional semi-arid wooded grassland in eastern Australia. Two grasses were palatable C3 species (Monachather paradoxa Steud. and Thyridolepis mitchelliana (Nees) S. T. Blake) and two were unpalatable C4 species (Aristida jerichoensis (Domin) Henr. var. subspinulifera Henr. and Eragrostis eriopoda Benth.). During the 10-year study the grasses were protected from large herbivore grazing within paddocks continuously grazed by sheep. Death occurred only during droughts and rates of death were species-dependent. When plotted against several water availability indices, rainfall and rainfall/evaporation during the preceding 3 months provided best predictions of death. Longer preceding periods gave inferior predictions. A 3-month rainfall total of 75 mm and a 3-month rainfall/evaporation ratio of 0.15 were survival critical thresholds below which deaths began. The 3-month rainfall totals, rainfall/evaporation and estimated water status of plants were equally reasonable predictors of deaths, but were inconsistent in their effectiveness. Rainfall was adopted for the grass death model; death begins when 3-month rainfall total declines below a threshold of 75 mm and the death rate rises with lower rainfall. Position of plants in the gently undulating landscapes influenced water status and, hence, death rates. Water status of grasses on the two water-shedding zones and the ‘flat’ zone were similar at each assessment, but higher on ‘ridge run-on’ and ‘toe-of-slope’ zones. Foliage height and diameter also influenced death rate but were species dependent. Basal diameter did not influence death rate. Survivorship of several perennial grass species at widely spaced sites in south-eastern Australia provided equivocal support for generality of the grass death model.