Countries that share rivers have a higher risk of initiating international militarized disputes, over and above being neighbors. They also have slightly higher levels of cooperation. Controlling for the length of the land boundary has shown that the shared river variable is not just a proxy for a higher degree of interaction opportunity. A weakness of earlier work is that the existing shared rivers data do not allow us to distinguish properly between dyads where the rivers run mainly across the boundary and dyads where the shared river runs along the boundary. Dyads with rivers running across the boundary should be expected to give rise to the traditional upstream-downstream scenario in the 'water wars' literature. Dyads where the river forms the boundary might be expected to be conflict-prone for a very different reason, because rivers boundaries are fluid and fuzzy. While the upstream-downstream type of shared river fits into to the neomalthusian resource scarcity scenario, the fuzzy boundary type does not. We test this with a new and improved dataset on shared rivers.