Although swarm-raiding army ants are considered keystone predators in tropical forest ecosystems, information on the prey spectra of most species is based on anecdotal reports and not on systematic studies with extensive sampling. We analysed prey samples of 18 colonies of the two afrotropical species Dorylus (Anomma) molestus and Dorylus (Anomma) wilverthi (4289 prey items in total) to examine the prey composition variation within and between species and to determine the best methodology to obtain reliable prey spectrum estimates for a given species, site and season. Variation in prey composition was substantial for D. molestus even within a single site and season, with the biomass proportion of the most important prey type differing by a factor of 12. Conclusions from studies using small samples sizes may thus be misleading. We demonstrate that the method of assessing prey spectra in terms of relative prey item numbers often produces biased results, and therefore recommend relative prey biomass as the more useful parameter. The near absence of earthworms, which always constituted a substantial part of the D. molestus prey, in the diet of D. wilverthi is interpreted to result from subtle differences in swarm-raiding behaviour between these two species, but could alternatively also be due to low availability. Similar studies recording prey composition as biomass proportions and analysing large samples sizes from many colonies are needed to understand the effect of army ant swarm raids on invertebrate communities in afrotropical forests.