The full nucleotide sequences of the chloroplast encoded large subunit of ribulose-1,5-bisphosphate carboxylase (rbcL) are available for nine grass species and partial sequence data for one species. Relative rate tests of the "molecular clock" hypothesis suggest that rbcL evolved more rapidly in the lineage leading to Zea than in those leading to the other species. The estimated overall substitution rate for rbcL among these grasses is about 5 times 10 substitutions per site per year, or about one-half the synonymous rate. The nine full sequences were analyzed by the UPGMA, Wagner parsimony, maximum likelihood, and Fitch-Margoliash methods. The latter three methods produced trees with the same topology. This topology largely agrees with current taxonomic evidence regarding the relationships among these grasses. UPGMA produced a topology that conflicts more substantially with available taxonomic evidence. Statistical comparison of the three alternative topologies for the subfamilies Panicoideae, Pooideae and Bambusoideae failed to support one of these topologies over the others, reflecting the taxonomic ambiguities surrounding the relationships among these taxa. Phylogenetic analyses based on the partial sequences of all 10 species gave conflicting results with regard to the relationship between Hordeum and Triticum, both members of the tribe Triticeae. This indicates that rbcL sequences contain too little information to resolve relationships among genera within this tribe. Overall, the results suggest that rbcL sequence data can provide some new information concerning grass phylogeny, but that the amount of available data from this gene is too small to differentiate statistically among alternative topologies for the grasses. Conflicting results from parsimony, maximum likelihood, and Fitch-Margoliash methods proved useful in exploring the validity of assumptions underlying these methods.