Abstract Motivation Where experiments identify sets of grass genes of unknown function, e.g. underlying a QTL or co-expressed in a transcriptome, it is useful to know which of these genes are common to all grasses (universal) and whether they likely have monocot-/commelinid-/grass-specific function. Results A pipeline used data on 16 grass full genomes from Ensembl Plants to generate 13 312 highly conserved, universal groups of grass protein-coding genes. Validation steps showed that 98.8% of these groups also had gene matches in recently sequenced genomes from two major grass clades not used in the pipeline. Comparison with many non-grass genomes identified 4609 of these groups as likely of monocot-/commelinid-/grass-specific function. Both grouping of genes and specificity were defined using hidden Markov model (HMM) profiles of the groups. The HMM-based approach performed better than simple percentage identity in discriminating between test sets of known specific and non-specific genes. The results give novel insight into the nature of monocot-/commelinid-/grass-specific genes. Researchers can use the universal_grass_peps database to gain evidence for their experimentally identified grass genes being involved in monocot-/commelinid-/grass-specific traits. Availability and implementation The universal_grass_peps database is available for download at https://data.rothamsted.ac.uk/dataset/universal_grass_peps.