The C4 Urochloa spp (syn. Brachiaria) and Megathyrsus maximus (syn. Panicum maximum) are used as pasture for cattle across vast areas in tropical agriculture systems in Africa and South America. A key target for variety improvement is forage quality: enhanced digestibility could decrease amount of land required per unit production and enhanced lipid content could decrease methane emissions from cattle. For these traits, loss-of-function (LOF) alleles in known gene targets are predicted to improve them, making a reverse genetics approach of allele mining feasible. We therefore set out to look for such alleles in a diverse Urochloa spp and Megathyrsus maximus accessions from the genebank collection held at CIAT.We studied allelic diversity of 20 target genes (11 for digestibility, 9 for lipid content) in 104 accessions selected to represent genetic diversity and ploidy levels of U. brizantha, U. decumbens, U. humidicola, U. ruziziensis and M. maximum. We used RNAseq and then bait-capture DNA-seq to improve gene models in a U. ruziziensis reference genome to assign polymorphisms with high confidence.We found 953 non-synonymous polymorphisms across all genes and accessions; within these, we identified 7 putative LOF alleles with high confidence, including ones in the non-redundant SDP1 and BAHD01 genes present in diploid and tetraploid accessions. These LOF alleles could respectively confer increased lipid content and digestibility if incorporated into a breeding programme.We demonstrated a novel, effective approach to allele discovery in diverse accessions using a draft reference genome from a single species. We used this to find gene variants in a collection of tropical grasses that could help reduce the environmental impact of cattle production.