Seed dispersal effectiveness (SDE) is the contribution of dispersers to plant recruitment and is estimated as the product of the number of seeds dispersed (quantity) and the probability of recruitment of each dispersed seed (quality). Although SDE is a key concept in seed dispersal ecology, few studies estimate SDE and none has a community approach. Oceanic islands, with simple communities, are ideal for this purpose. In this study, we compared the SDE of the main types of dispersers (lizards and passerine birds) at the community level in a given habitat. We estimated SDE using a stochastic simulation model parameterized with empirical data on quantity and quality components measured throughout the recruitment process. Although lizards are highly frugivorous and their density was approximately 20 times higher than that of birds, lizards and birds dispersed a similar quantity of seeds. This may be due to lower intake of seeds by lizards due to their slower metabolism (approximately 20 times lower than birds). This low metabolic rate limits the importance of lizards as seed dispersers, but it is compensated by extraordinarily high lizard densities in the study area (approximately 9600 individuals/km2). High densities of lizards are typical of islands, and this helps to explain why dispersal by lizards seems mainly an island phenomenon. Birds and lizards showed functional complementarity, especially regarding seed dispersal distribution patterns. In fact, lizards dispersed more seeds in shrublands and open sites, and birds in woodlands and beneath canopies, with their joint contribution helping to maximize recruitment. Lizards provided higher SDE than birds for 7 out of 11 plant species. The disperser with a higher quantity for a given plant generally had the higher quality, and plants could be classified as bird- or lizard-dependent for dispersal. This dependence increased when considering SDE instead of dispersal quantity only. Moreover, quality was a better predictor of SDE than quantity, which should be considered when parameterizing interaction networks, as this might affect inferences about their architecture.