Local sampling of larval lepidopterans on Erythroxylum host plant species in tropical savanna (cerrado) revealed a high species richness with low abundance per species. Cumulative numbers of morphospecies with increasing sampling effort yielded no asymptotic level of richness in sampling periods lasting 6 mo, 7 mo, and 23 mo. Peak richness was reached at 31 species in 1992 and 19 species in 1993, on the three Erythroxylum species sampled: E. deriduum, E. suberosum, and E. tortuosum. Less than one larva was found per plane during all sampling times, with a mean of 0.28 species per plant in 1992 and 0.12 in 1993. The number of specimens of all species combined discovered per plant was very low at 0.10 per plant. Many plants remained unattacked on any sampling date with 12 percent, 8 percent, and 16 percent of planes attacked in the species listed in order above. In general, there was an increase in species found during the late dry season when new leaves were produced, but plane phenology seemed to exert only a small influence. Comparisons with temperate samples of a similar kind, in savanna vegetation at the same altitude, indicate a very different assemblage. Comparison was based on four criteria: richness was from two to over three times higher in the tropics, even though sampling had not produced an asymptotic accumulation of species; the number of morphospecies per plant individual was similar at the sites, although total richness was lower in the temperate savanna; the number of total individuals per plant was 11-fold higher in the temperate samples; and the percent of plants with larvae present was over four times higher in the temperate zone (mean of 49%) than in the cerrado (12%). The high richness of relatively rare species in the cerrado site poses challenges in understanding the reasons for such rarity, the organization of such assemblages, the gradient of species richness from low to high latitudes, the estimation of biodiversity, and conservation management.