In environmental metagenomic experiments, a very high proportion of the microbial sequencing data (> 70%) remains largely unexploited because rare and closely related genomes are missed in short-read assemblies. The identity and the potential metabolisms of a large fraction of natural microbial communities thus remain inaccessible to researchers. The purpose of this study was to explore the genomic content of unassembled metagenomic data and test their level of novelty. We used data from a three-year microbial metagenomic time series of the NW Mediterranean Sea, and conducted reference-free and database-guided analysis. The results revealed a significant genomic difference between the assembled and unassembled reads. The unassembled reads had a lower mean identity against public databases, and fewer metabolic pathways could be reconstructed. In addition, the unassembled fraction presented a clear temporal pattern, unlike the assembled ones, and a specific community composition that was similar to the rare communities defined by metabarcoding using the 16S rRNA gene. The rare gene pool was characterised by keystone bacterial taxa, and the presence of viruses, suggesting that viral lysis could maintain some taxa in a state of rarity. Our study demonstrates that unassembled metagenomic data can provide important information on the structure and functioning of microbial communities.