Melastomataceae sensu stricto (excluding Memecylaceae) comprise some 3000 species in the neotropics, 1000 in Asia, 240 in Africa, and 230 in Madagascar. Previous family–wide morphological and DNA analyses have shown that the Madagascan species belong to at least three unrelated lineages, which were hypothesized to have arrived by trans–oceanic dispersal. An alternative hypothesis posits that the ancestors of Madagascan, as well as Indian, Melastomataceae arrived from Africa in the Late Cretaceous. This study tests these hypotheses in a Bayesian framework, using three combined sequence datasets analysed under a relaxed clock and simultaneously calibrated with fossils, some not previously used. The new fossil calibration comes from a re–dated possibly Middle or Upper Eocene Brazilian fossil of Melastomeae. Tectonic events were also tentatively used as constraints because of concerns that some of the family's fossils are difficult to assign to nodes in the phylogeny. Regardless of how the data were calibrated, the estimated divergence times of Madagascan and Indian lineages were too young for Cretaceous explanations to hold. This was true even of the oldest ages within the 95% credibility interval around each estimate. Madagascar's Melastomeae appear to have arrived from Africa during the Miocene. Medinilla, with some 70 species in Madagascar and two in Africa, too, arrived during the Miocene, but from Asia. Gravesia, with 100 species in Madagascar and four in east and west Africa, also appears to date to the Miocene, but its monophyly has not been tested. The study afforded an opportunity to compare divergence time estimates obtained earlier with strict clocks and single calibrations, with estimates based on relaxed clocks and different multiple calibrations and taxon sampling.