Species distribution models (SDMs) are one of the most widely used tools to predict areas with potential for occurrence of native, invasive and endangered species, based on current and future environmenal and climate conditions. Despite their global use, evaluating the accuracy of SDMs based only on presence records is still a challenge. The performance of models depends on the sample size and species prevalence. Recently, studies to model the distribution of species in the Caatinga biome in Northeast Brazil have gained force, raising the question about the minimum number of presence records adjusted to different prevalences that are necessary to generate accurate SDMs. In this context, the objective of this study was to indicate minimum numbers of presence records for species with different prevalences in the Caatinga biome to obtain accurate SDMs. For that purpose, we used a method involving simulated species and performed repeated evaluations of the models' performance in function of the sample size and prevalence. The results indicated that for this approach in the Caatinga biome, the minimum required numbers of specimen records were 17 and 30 for species with narrow and widespread distributions, respectively.