JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.. British Ecological Society is collaborating with JSTOR to digitize, preserve and extend access to Journal of Ecology. Summary 1 Two recently published models of succession are the gradient-in-time model and the competitive-sorting model. The latter predicts that community structure will change during the course of succession, with chance establishment predominating early in succession, and competitive sorting into microhabitats increasing with time. The average strength of positive and negative interspecific associations should therefore increase. There is little evidence available to test these models in real communities. 2 We test the two models using postfire succession in tall tussock grasslands of New Zealand. Twenty-one sites dominated by the tussock grass Chionochloa rigida, and last burnt within the past 30 years, were sampled using contiguous quadrats. Three association coefficients were calculated for each site and related to time since the last fire, presence or absence of grazing, maximum height of vegetation and soil nutrient status. The results were examined at a range of spatial scales. 3 Community structure, as evidenced by the strength of interspecific association at small quadrat sizes, was significantly related to site factors. Mean association was negative in recently burnt sites, and again in sites burnt more than 20 years previously. The trends support the competitive-sorting model over the gradient-in-time model. The results are also compatible with the 'three-phase' model of secondary succession suggested by Greig-Smith; we refer to the three phases as 'pioneer ', 'building' and 'mature'. 4 There was a significant tendency over all sites for there to be a constant proportion of species from the graminoid vs. forb + bryophyte + lichen guilds in a quadrat, compared to null-model expectation. This tendency towards guild proportionality was more marked in lower elevation sites, and in sites where there was no grazing. This supports the existence of determinate community structure, a factor explicitly included in the competitive-sorting model. 5 We conclude that there is significant community structure in the tussock grasslands. The changes in species association are affected by grazing and site quality. The strength of species association can be an index of changes in community structure during succession, in spite of the confounding effect of microsite variation; however, this relationship is affected by grazing and site quality. The competitive-sorting model fits the results more closely than the gradient-in-time model, but the 'three-phase' model gives the best fit.