We describe a phylogeny of the Bovidae based on 40 allozyme loci in 27 species, representing 10 of the 14 bovid tribes described by Vrba (1985). Giraffe represented a related family (Giraffidae). A phenogram was derived using the unweighted pair-group method with arithmetic means (UPGMA), based on Nei's genetic distances (ND) between species. A tree was also derived using the neighbor-joining technique, also based on ND. To provide a cladistic interpretation, the data were analyzed by a maximum parsimony method (phylogenetic analysis using parsimony, PAUP). We found marked divergence within the Bovidae, consistent with the appearance of the family in the early Miocene. Unexpectedly, the most divergent species was the impala, which occupied a basal position in all trees. Species in the tribe Alcelaphini were the most derived taxa in all trees. These patterns conflict strongly with the previous taxonomic alliance, based on immuno-distance and anatomical evidence, of the impala as a sister group of the Alcelaphini. All trees agreed that tribes described by Vrba (1985) are monophyletic, except the Neotragini, which was polyphyletic, with suni occupying a long branch by itself. The dikdik and klipspringer were consistently placed as sister taxa to species in the Antilopini. Three tribes (Aepycerotini, Tragelaphini and Cephalophini), whose fossils have not been found outside Africa, were basal in all trees, suggesting that bovids originated in Africa. Nodes connecting the remaining tribes were closely clustered, a pattern that agrees with fossil evidence of rapid divergence within the Bovidae in the mid-Miocene (about 15 mybp). The allozyme data suggested a second phase of rapid divergence within tribes during the Plio-Pleistocene, a pattern that also agrees with fossil evidence. Rates of bovid divergence have therefore been far from constant. However, the clustering of nodes imparts considerable uncertainty to the branching order leading to the derived tribes, and to a lesser extent, species within tribes. The classical division of the Bovidae into the Boodontia and Aegeodontia does not agree with the phylogenetic grouping of tribes presented in this analysis. However, the maximum parsimony tree derived using 'local' branch swapping clustered all grazing species into a derived, monophyletic group, suggesting that grazing may have evolved only once in bovid evolution.