Species richness varies enormously across geographical gradients, a well-known phenomenon for which there are many hypothesized explanations. One recent hypothesis uses null models to demonstrate that random re-distribution of species’ ranges within a given domain leads to a ‘mid-domain effect’ (MDE): increasing species richness towards the centre of the area. Madagascar is especially well-suited for empirical evaluation of mid-domain models by virtue of its large endemic fauna and its clearly defined boundaries. Lees et al. [Biol. J. Linn. Soc.67 (1999) 529] observed patterns of species richness consistent with MDEs in the Madagascan rainforest (a slim, north–south belt). In this study, we test one-dimensional and two-dimensional mid-domain model predictions for the birds and mammals of the entire island of Madagascar. When only latitudinal extents of species’ distribution are considered, patterns of richness in Madagascar show an MDE. However, this pattern disappears for both taxa after accounting for the tendency of latitudinal bands nearer the middle of the country to be larger. Two-dimensional mid-domain model predictions of species richness are qualitatively opposite to observed patterns. Instead, island-wide spatial gradients of species richness in Madagascar relate strongly to patterns of primary productivity and amount of remaining natural habitat. Earlier work that showed a mid-domain peak within the rainforest biome (effectively after controlling for climate and natural habitat) seems likely to have reflected methodological artefacts. The classic case in which MDEs should occur is, in fact, inconsistent with the mid-domain hypothesis.