Many decisions for pest control in agriculture are based on densities, rather than species presence only. Little is known about the distribution of pest abundance and even less when several species present in a field may interact. With the aim of developing an ecologically based approach to weed management, we compared observed co-occurrence patterns with a null model, to test whether the abundances of weed species were randomly assembled. The data set covered the pattern of co-occurrence of abundances of weeds within 200-m2 plots located in 1143 arable fields scattered across France. Our results pinpointed a highly significant _similar attraction–dissimilar repulsion_rule, as species reaching similar abundance levels co-occurred significantly more often than expected under the hypothesis of a random distribution of abundance. A similar analysis applied to the 25% plots that had the highest weed densities indicated that the _similar attraction–dissimilar repulsion_ rule held true for low-abundance classes, but not in situations where at least one species was observed at densities above 20 individuals m)2. Our analysis extends the analysis of community beyond the sole species richness index by accounting for their abundance. Overall, weed species interfered or responded similarly to an external biotic or abiotic factor, with resulting adjustment of densities at the scale of the agricultural field