The strong link between environment and the ecological diversity of communities is often used for drawing palaeoenvironmental inferences from fossil assemblages. Here we focus on the reliability of fossil samples in comparison to original communities when inferring palaeoenvironments from the ecological diversity of fossil mammal faunas. Taphonomic processes and sampling techniques generally introduce two kinds of biases in fossil samples: 1) the directional impoverishment of communities, i.e. the absence of some specific categories of bones, individuals or species; and 2) the mixture of several communities, temporally (timeaveraging) and/or spatially (space-averaging). We explore the impact of such alterations on ecological diversity and taxonomic richness by simulating an impoverishment in small species (1-45 kg) and a mixture of communities on an original dataset of 50 modern African communities. The progressive impoverishment in small species in a mammal community induces a relative enrichment in terrestrial and grazing species and a depletion (or even the disappearance) of several ecological categories (e.g., fossorial, arboreal, sub-arboreal, frugivorous, omnivorous and insectivorous species), thus leading to the ecological homogenisation of the altered communities. Other categories (carnivorous species, browsers, mixed feeders and sub-aquatic species) prove stable and seem to be good estimates for the relative abundances in the parent communities. Ideally, palaeoenvironmental inferences should be drawn from those ecological categories. For strong degrees of impoverishment, the discriminating power of the ecological diversity is reduced and other proxies should be used. Once the degree of impoverishment is assessed, it becomes possible to consider the condensation of several communities into a single fossil sample by comparing its taxonomic richness with those of modern communities, first impoverished to the same level. The mixture of communities tends to increase the taxonomic richness but does not significantly modify the overall ecological diversity. As a case study, we applied this approach to 15 African mammal samples from the late Miocene-earliest Pliocene, a period of particular interest, as it saw the emergence of hominins, together with faunal turnovers, and major climatic and environmental changes. Most of our fossil samples, especially those that yielded early hominins, appear altered to different degrees by taphonomic processes and sampling techniques, meaning that their faunal structures may not be comparable and that their inferred palaeoenvironmental differences may not be genuine.