China's rice fields are the oldest known anthropogenic wetland habitats. They are home to diverse weed communities that have been shaped over millennia by hydrological conditions, soil properties, and crop development. Here, we present the results of research on rice field weed vegetation, using 167 phytosociological relevés (10 m²) collected in southern China in 2016, supplemented with soil samples from 69 of these plots. The vegetation was classified using a modified TWINSPAN method, while relationships with environmental variables were analysed using CCA with interactive forward selection. Floristic distances between clusters were additionally assessed using 2D NMDS and mean Bray–Curtis dissimilarities. A total of 183 taxa (130 genera, 52 families) were recorded in the entire dataset, and the TWINSPAN method allowed nine vegetation clusters to be identified, including two pleustonic aquatic units (Ricciocarpetum natantis and Azolletum pinnatae) and several Oryzetea rice-weed communities (including Rotaletum rotundifoliae and Hyperico monanthemi–Ammannietum bacciferae). In the CCA analysis, significant factors differentiating the communities included precipitation seasonality (BIO15), annual precipitation (BIO12), altitude above sea level, water depth, soil type and texture fractions (>2 mm; <0.002 mm), soil pH (H2O), and rice height. The selected CCA model explained 39.7% of the total variation in composition (19.5% after adjustment). Pleustonic water clusters showed exceptionally high dissimilarity to most rice weed clusters, while several rice field communities showed comparatively lower, but still substantial, floristic dissimilarity. Our results show that rice agroecosystems in southern China are characterised by high diversity, associated with geographic, hydrological, and edaphic gradients, a finding relevant to biodiversity conservation in traditional rice landscapes and weed management.