Point pattern analysis (fitting of the beta-binomial distribution and binary form of power law) was used to describe the spatial pattern of natural take-all epidemics (caused by Gaeumannomyces graminis var. tritici) on a second consecutive crop of winter wheat in plots under different cropping practices that could have an impact on the quantity and spatial distribution of primary inoculum, and on the spread of the disease. The spatial pattern of take-all was aggregated in 48% of the datasets when disease incidence was assessed at the plant level and in 83% when it was assessed at the root level. Clusters of diseased roots were in general less than 1 m in diameter for crown roots and 1–1·5 m for seminal roots; when present, clusters of diseased plants were 2–2·5 m in diameter. Anisotropy of the spatial pattern was detected and could be linked to soil cultivation. Clusters did not increase in size over the cropping season, but increased spatial heterogeneity of the disease level was observed, corresponding to local disease amplification within clusters. The relative influences of autonomous spread and inoculum dispersal on the size and shape of clusters are discussed.