River networks play a key role in the spatial organization of human settlements. Both river networks and human settlements have been found to exhibit regular self-similar patterns, but little is known about the generalized spatial patterns of human settlements embedded within river networks. Here based on night light data, we find a universal fractal structure at the global scale, with both robust Hortonian scaling relationships with the extent of human settlements and statistically significant power law scaling of the power spectra of human area functions. Globally, we find consistent patterns of power law preferential downstream clustering of human settlements across all six populated continents, typically up to 40% of the maximum flow length. This downstream clustering suggests an optimum distribution of humans in large river basins for trade, transport, and natural resource utilization but with attendant implications for human impacts on rivers. Recognition of such spatial patterns helps generalize assessments of human impacts on rivers, with direct implications for management of water quality and biological diversity in river networks.