1. The unpredictable component of community structure – chance/randomness/stochasticity – is increasingly discussed, recently in terms of Neutral Theory, but is never directly measured. We show that a direct estimate of chance can be made from spatial autocorrelation, and demonstrate this usage.
2. The dissimilarity between two patches of vegetation was plotted against the distance between the patches, and a curve fitted. We argue that the y-intercept of the curve, i.e. the ‘nugget’, represents the amount of chance variation in species composition, because when distance is zero all dispersal limitation and spatially correlated environmental differences are excluded, and only chance remains.
3. The method estimated that in 16 sites around the South Island of New Zealand, 0–71% of the variation in plant community composition was due to chance, with a mean of 34%.
4. Synthesis. The new analysis method provides information on the amount of randomness in community species composition and could be applied to the many dissimilarity/distance studies in the literature. The amount of randomness in the 16 study sites was not related to the type of plant community or to the species richness per quadrat. However, randomness was positively correlated with whole-community species richness, supporting previous suggestions of a relationship between chance, the size of the species pool and redundancy.