A comprehensive understanding of the effects of tillage practices on soil microbial communities is fundamental to a better understanding of their roles in maintaining or improving the stability of agroecosystems. However, little is known regarding soil microbial co-occurrence patterns and their function in ecosystems shaped by long-term tillage practices. The goal of the present study was to investigate the impact of no-tillage (NT), reduced tillage (RT) and moldboard plow tillage (MP) practices on soil microbial interactions, central species and functional taxa as well as to investigate their associations with crop yields. The results showed that the average node degree was lower in the RT network than in the NT and MP networks of bulk and rhizosphere soils, indicating inefficient mutual effects among taxa and a more stable network structure induced by RT. Moldboard plow tillage resulted in the most fungal-fungal interactions (17.6%), while RT yielded the most bacterial-fungal interactions (32.3%) in the rhizosphere soils. The potential plant pathogens Fusarium were identified as hub node with most connections, and pathogen Sphingomonas and Clonostachys were found as central speices with those simultaneously termed as hub node and generalist in the MP network of rhizosphere soils. Intriguingly, these potential plant pathogens had negative correlations with crop yields in both bulk and rhizosphere soils, as determined by structural equation modeling (SEM). In contrast, RT might help to alleviate the ecological risks of the potential plant pathogens and increased the availability of soil nutrients mediated by central species, such as Nitrospira. These findings suggested RT as an optimal practice that could stabilize microbial network structure, alleviated the potential transmission of pathogens and finally enhance crop yields, providing a new perspective for tillage application in agroecosystems.