DNA sequence data from the chloroplast gene ndhF were analyzed to estimate the phylogeny of the subfamily Panicoideae, with emphasis on the tribe Paniceae. Our data suggest that the subfamily is divided into three strongly supported clades, corresponding to groups with largely identical base chromosome numbers. Relationships among the three clades are unclear. In unweighted parsimony analyses, the two major clades with x = 10 (Andropogoneae and x = 10 Paniceae) are weakly supported as sister taxa. The third large clade corresponds to x = 9 Paniceae. In analyses under implied weight, the two clades of Paniceae are sisters, making the tribe monophyletic. Neither resolution is strongly supported.Our molecular phylogenies are not congruent with previous classifications of tribes or subtribes. Based on this sample of species, we infer that C(4) photosynthesis has evolved independently several times, although a single origin with multiple reversals and several reacquisitions is only slightly less parsimonious. The phosphoenol pyruvate carboxykinase (PCK) subtype of C(4) photosynthesis has evolved only once, as has the NAD-malic enzyme (ME) subtype; all other origins are NADP-ME. Inflorescence bristles are apparently homologous in the genera Setaria and Pennisetum, contrary to opinions of most previous authors. Some genera, such as Digitaria, Echinochloa, and Homolepis are supported as monophyletic. The large genus Paspalum is shown to be paraphyletic, with Thrasya derived from within it. As expected, Panicum is polyphyletic, with lineages derived from multiple ancestors across the tree. Panicum subg. Panicum is monophyletic. Panicum subg. Dichanthelium, subg. Agrostoides, and subg. Phanopyrum are unrelated to each other, and none is monophyletic. Only Panicum subg. Dichanthelium sect. Dichanthelium, represented by P. sabulorum and P. koolauense, is monophyletic. Panicum subg. Megathyrsus, a monotypic subgenus including only the species P. maximum, is better placed in Urochloa, as suggested by other authors.