A theory of spatial structure in ecological communities is presented and tested. At the core of the theory is a simple allocation rule for the assembly of species in space. The theory leads, with no adjustable parameters, to nonrandorn statistical predictions for the spatial distribution of species at multiple spatial scales. The distributions are such that the abundance of a species at the largest measured scale uniquely determines the spatial-abundance distribution of the individuals of that species at smaller spatial scales. The shape of the species-area relationship, the endemics-area relationship, a scale-dependent community-level spatial-abundance distribution, the species-abundance distribution at small spatial scales, an index of intraspecific aggregation, the range-area relationship, and the dependence of species turnover on interpatch distance and on patch size are also uniquely predicted as a function solely of the list of abundances of the species at the largest spatial scale. We show that the spatial structure of three spatially explicit vegetation census data sets (i.e., a 64-m(2) serpentine grassland plot, a 50-ha moist. tropical forest plot, and a 9.68-ha dry tropical forest plot) are generally consistent with the predictions of the theory, despite the very simple statistical assumption upon which the theory is based, and the absence of adjustable parameters. However, deviations between predicted and observed distributions do arise for the species with the highest abundances; the pattern of those deviations indicates that the theory, which currently contains no explicit description of interaction mechanisms among individuals within species, could be improved with the incorporation of intraspecific density dependence.