Abstract Background Efficient industrial processes for converting plant lignocellulosic materials into biofuels are a key challenge in global efforts to use alternative energy sources to fossil fuels. Novel cellulolytic enzymes have been discovered from microbial genomes and metagenomes of microbial communities. However, the identification of relevant genes without known homologs, and elucidation of the lignocellulolytic pathways and protein complexes for different microorganisms remain a challenge. Results We describe a new computational method for the targeted discovery of functional modules of plant biomass-degrading protein families based on their co-occurrence patterns across genomes and metagenome datasets, and the strength of association of these modules with the genomes of known degraders. From more than 6.4 million family annotations for 2884 microbial genomes and 332 taxonomic bins from 18 metagenomes, we identified five functional modules that are distinctive for plant biomass degraders, which we call plant biomass degradation modules (PDMs). These modules incorporated protein families involved in the degradation of cellulose, hemicelluloses and pectins, structural components of the cellulosome and additional families with potential functions in plant biomass degradation. The PDMs could be linked to 81 gene clusters in genomes of known lignocellulose degraders, including previously described clusters of lignocellulolytic genes. On average, 70% of the families of each PDM mapped to gene clusters in known degraders, which served as an additional confirmation of their functional relationships. The presence of a PDM in a genome or taxonomic metagenome bin allowed us to predict an organism’s ability for plant biomass degradation accurately. For 15 draft genomes of a cow rumen metagenome, we validated by cross-linking with confirmed cellulolytic enzymes that the PDMs identified plant biomass degraders within a complex microbial community. Conclusions Functional modules of protein families that realize different aspects of plant cell wall degradation can be inferred from co-occurrence patterns across (meta)genomes with a probabilistic topic model. The PDMs represent a new resource of protein families and candidate genes implicated in microbial plant biomass degradation. They can be used to predict the ability to degrade plant biomass for a genome or taxonomic bin. The method would also be suitable for characterizing other microbial phenotypes.