The milpa is a traditional maize-based polyculture in Mexico that is typically practiced as rainfed agriculture. Because milpa cultivation has been practiced over a vast range of environmental and cultural conditions, this agroecosystem is recognized as an important repository of biological and cultural diversity. As for any agroecosystem, the relationship between plant development and the biogeochemical processes of the soil is critical. Although the milpa has been studied from different perspectives, the diversity and structure of microbial communities within milpa soils remain largely unexplored. In this study, we surveyed a milpa system in Central Mexico across cropping season: before planting (dry season; t1), during the early growth of plants (onset of the rainy season; t2), and before harvest (end of the rainy season; t3). In order to examine changes in community structure through time, we characterized bacterial diversity through high-throughput sequencing of 16S rRNA gene amplicons and recorded the nutrient status of multiple (5–10) soil samples from our milpa plots. We estimated microbial diversity from a total of 90 samples and constructed co-occurrence networks. Although we did not find significant changes in diversity or composition of bacterial communities across time, we identified significant rearrangements in their co-occurrence network structure. We found particularly drastic changes between the first and second time points. Co-occurrence analyses showed that the bacterial community changed from a less structured network at (t1) into modules with a non-random composition of taxonomic groups at (t2). We conclude that changes in bacterial communities undetected by standard diversity analyses can become evident when performing co-occurrence network analyses. We also postulate possible functional associations among keystone groups suggested by biogeochemical processes. This study represents the first contribution on soil microbial diversity of a maize-based polyculture and shows its dynamic nature in short-term scales.