Recent theoretical and empirical research suggests that statistical models based on coalescent theory can improve both phylogeographic and phylogenetic inference. An approach that involves elements of both statistical phylogeography (e.g. Isolation with Migration analyses) and multilocus phylogenetic inference (e.g. *beast) may be particularly useful when applied to populations with relatively old divergence times. Here, we use such an approach in the globally distributed brown booby (Sula leucogaster). We sampled 215 individuals from all major breeding areas and genotyped them at eight microsatellite and three nuclear intron loci. We found that brown booby populations were highly differentiated and that colonies can be grouped into four major genetic populations (Caribbean Sea, Central Atlantic Ocean, Indo-Central Pacific and Eastern Pacific). These populations apparently diverged in the absence of gene flow and, with one exception, currently exchange few to no migrants. The Eastern Pacific population diverged from all other populations approximately one million years ago [90% highest posterior density: 330 000–2 000 000 years ago] and exhibits a distinct male plumage, relative to other populations. However, recent gene flow from the Indo-Central Pacific into the Eastern Pacific appears to have occurred, suggesting that approximately one million years of genetic isolation and divergence in male plumage colour are not sufficient to prevent interbreeding. Gene flow following secondary contact of the Indo-Central Pacific and Eastern Pacific populations was not detected in previous mitochondrial DNA (mtDNA) studies, and the contrast between the mtDNA results and our current results highlights the advantage of a multilocus phylogeographic approach.