On 26 December 2004, the Mw 9.2 Sumatra earthquake triggered the greatest tsunami ever observed in recent times. The unprecedented quality and variety of observations on modern instruments provide opportunities to refine modeling techniques as well as our perception of tsunami hazard in oceanic regions where historical events are not as frequent as in the Pacific Ocean. We present a numerical modeling of the 2004 tsunami on the Mascarene Islands with refined computations on La Réunion Island, where postevent survey observations and one tide gauge record are available. We use two different source models (models A and B) to account for the observations, and we also test an additional source model (model C) to study the effect of a tsunami originating in the 1833 epicentral area (southern Sumatra). A wave trapped on the edge of La Réunion may explain the greatest water heights reached in the western coastline of the island. The observed tide gauge data are reasonably well fit (models A and B). The run-up data can also be explained with our models. Late disturbances observed in La Pointe des Galets may be explained by a late tsunami reflection from Madagascar or by harbor resonances. Finally we also show that model C (1833-like) strikes La Réunion more heavily than models A and B.