African Great Lake cichlid populations are divided into thousands of genetic subpopulations. The low gene flow between these subpopulations is thought to result from high degrees of natal philopatry, heavy predation pressure, and a patchy distribution of preferred habitats. While predation pressure and habitat distribution are fairly straightforward to assess, data on dispersal distances and rates are scarce. In fishes, direct observations of dispersal events are unlikely, but dispersal can be studied using molecular markers. Using seven microsatellite loci, we examined dispersal in the cooperatively breeding cichlid fish, Neolamprologus pulcher. As this species is found in well-defined groups clustered into subpopulations, we could assess dispersal on a narrow (within subpopulation) and broad (between subpopulation) scale. While fish were generally more related to others in their own subpopulation than they were to fish from other subpopulations, large males diverged from this pattern. Large males were more related to other large males from different subpopulations than they were to large males from their own subpopulation, suggesting more frequent dispersal by large males. Across subpopulations, relatedness between large males was higher than the relatedness among large females; this pattern was not detected in small males and small females. Within a subpopulation, individuals appeared to be preferentially moving away from relatives, and movement was unrestricted by the physical distance between groups. Our results highlight the importance of examining multiple spatial scales when studying individual dispersal biases.