Spatial and seasonal patterns of seed removal by rodents, ants and birds were evaluated using seed-dish experiments in a Trachypogon savanna in Venezuela. Four-day experiments were conducted on three dates and at four different sites. Seeds from 10 native plant species representing a broad mass range were offered at densities equivalent to 3145 seeds m−2. We found that total seed removal and relative importance of granivorous groups varied widely among sites and dates. Rodents were the major seed predators in all the habitats, except for the savanna/forest border where birds dominated. Ants removed more seeds than birds, but since they preferred the smallest seeds, ants ranked last in mass removed. We found a temporally variable relationship between seed removal rates and seed rain or seed bank. We inferred monthly seed predation from seed-dish experiments as well as through the decline of the seed bank in the environment during the dry season, and compared both estimations to evaluate the suitability of the seed-dish technique for estimating granivory. We initially hypothesized that seed-dish experiments that emulate seed availability more closely may yield a better estimate of seed predation, but in fact, they may still overestimate granivory intensity.