We used high-throughput sequencing and multivariate analyses to describe soil microbial community composition in two four-year field plant–soil feedback (PSF) experiments in Minnesota, USA and Jena, Germany. In descending order of variation explained, microbial community composition differed between the two study sites, among years, between bulk and rhizosphere soils, and among rhizosphere soils cultivated by different plant species. To try to identify soil organisms or communities that may cause PSF, we correlated plant growth responses with the microbial community composition associated with different plants. We found that plant biomass was correlated with values on two multivariate axes. These multivariate axes weighted dozens of soil organisms, suggesting that PSF was not caused by individual pathogens or symbionts but instead was caused by 'many weak' plant–microbe interactions. Taken together, the results suggest that PSFs result from complex interactions that occur within the context of a much larger soil microbial community whose composition is determined by factors associated with 'site' or year, such as soil pH, soil type, and weather. The results suggest that PSFs may be highly variable and difficult to reproduce because they result from complex interactions that occur in the context of a larger soil microbial community.