This study examined whether the response of herbaceous vegetation to environmental disturbances such as mechanical soil disturbance and fertilization can be predicted from a knowledge of the component species' plant attributes. Some aspects of Grime's (1979) C-S-R model were also tested. The model proposed that the three permutations of low stress and low disturbance, low stress and high disturbance, and high stress and low disturbance are associated with competitive (C), ruderal (R) and stress-tolerant (S) plant strategies, respectively. A completely randomised block experiment with three factors, soil disturbance, fertilization and sowing, was established in an area of moderately grazed temperate grassland on the Northern Tablelands of New South Wales. The site was dominated by the tall, warm season perennial grasses, 'Aristida ramosa' and 'A. warburgii' and its yellow podzolic soil was deficient in phosphorus, nitrogen and sulphur. The study area experiences an annual average rainfall of 914 mm, an annual average maximum temperature of 20.3°C and an annual average minimum temperature of 7.3°C. There were three levels of soil disturbance (none, hand-hoed and rotary-hoed), two levels of fertilization (none and fertilized NPS) and two levels of sowing (none and sown). The sown treatment consisted of hand-sowing the propagules of seven exotic species into the plots. During the experimental period (August 1992 -February 1994), the sown treatment was applied once in August 1992, disturbance was applied annually in August, fertilizer was applied twice a year and the vegetation was sampled in spring (November) and summer (March/February) each year. ... In conclusion, the response of herbaceous vegetation to mechanical soil disturbance and fertilization could be predicted from a knowledge of the component species' plant attributes. Plant attributes relating to plant persistence, biomass partitioning (leaves vs reproduction vs storage), and use of resources (conservative versus liberal) were strongly related to fertilization and soil disturbance. Plant attributes can discriminate between habitats as well as, or better than, floristic data. On a community scale, the influence of historical factors on vegetation response was not removed by the plant attribute approach, although it was somewhat reduced. Thus, the relationships found using this approach were of intermediate generality and are applicable to sites with a similar fertilizer and disturbance history. The greatest support for the C-S-R model was found using the key (4 predictions upheld, 1 inconsistent and 2 rejected). Support for the model was also found by testing its underlying hypotheses (2 predictions upheld, 2 inconsistent and 2 rejected), while the ordination of vegetation samples by the mean Rmax and MI of its component species was unsuccessful (0 predictions upheld, 2 inconsistent and 3 rejected). The present study shows how vegetation models may be viewed as a starting point for research and refinement rather than something that must be either wholly accepted or rejected.