1. Communities of different species are often structured according to niche differentiation associated with competitive interactions. We show that similar principles may apply on an ecological time-scale when individuals of a species having a wide size variation compete for resources, using the Lake Tanganyika cichlid Lobochilotes labiatus (5-30 cm). This species has a mouth especially adapted to suck up invertebrates from rock crevices. 2. Individuals defended feeding territories against similar-sized conspecifics, but not against different-sized ones. Thus, territories of similar-sized fish rarely overlapped, but up to a total of seven individuals (of seven size-ranks) had broadly overlapping territories with dissimilar-sized individuals. Comparison with expectation from the null model demonstrated clearly that observed size ratios between adjacent size rank were determined non-randomly regardless of sexual combinations. 3. Larger individuals took larger prey types of larger average size, but more importantly used wider rock crevices from which to suck food than smaller individuals. We calculated pairwise values of Schoener's index of diet overlap C(d) and the values of Levin's index of diet breadth B(d) (prey type and prey size) and the same for the width of the rock crevices used for foraging (C(r) and B(r)). C(d) remained high among all combinations of the seven ranks. In contrast, C(r) declined strongly in combinations of adjacent ranks (to 0.27), and was low or zero among further different size ranks. This shows that fish with overlapping territories divided the food resources largely through foraging site partitioning. Accordingly, B(d) did not depend on the size difference to the nearest two coinhabiting fish, whereas B(r) did. 4. We conclude that this L. labiatus community is structured non-randomly: body size-dependent effects on foraging site usage result in competition with, and territorial exclusion of, similar-sized individuals, but not of dissimilar-sized individuals that were accepted as coinhabitants. Accordingly, mean body size ratios (large/small) between two adjacent ranks were consistently approximately 1.28 [standard deviation (SD) = 0.07, n = 104], while approximately 1.34 from the null model (SD = 0.34, n = 10 400 simulations). We discuss our results as an example of Hutchinson's rule, applied originally to size ratios of different species.