The present study aimed to understand how assembly mechanisms drive the global and regional patterns of fish assemblage composition in estuaries. The approach used applied a bootstrapped hierarchical cluster analysis based on pairwise beta-dissimilarities (Beta sim) of fish assemblages between 393 estuaries to define biogeographical units. A set of large and small-scale filters were then used to test their influence on beta-dissimilarity patterns, through distance-based linear models (DISTLM). The global pattern obtained (i.e. seven major biogeographical units) was explained by large-scale filters related with geographic barriers (e.g. Isthmus of Panama) and temperature/current filters, in addition to their evolutionary history. Meanwhile, species composition within each biogeographical unit was also determined by large-scale filters, with only a minor influence of a few small-scale filters (i.e. tide range, estuary type and estuary area). Overall, the global pattern of fish composition in estuaries was mainly driven by dispersal limitation assembly mechanism (i.e. evolutionary history and geographical barriers/filters). In contrast with known species richness patterns, results support a weak influence of environmental filtering on species composition at regional scales, which was also driven by dispersal limitation. Results suggest a hierarchical influence of environmental filtering mechanism that acts at increasingly finer scales.