Bell inequalities bound the strength of classical correlations between observers measuring on a shared physical system. However, studies of physical correlations can be considered beyond the standard Bell scenario by networks of observers sharing some configuration of many independent physical systems. Here, we show how to construct Bell-type inequalities for correlations arising in any tree-structured network i.e. networks without cycles. This is achieved by an iteration procedure that in each step allows one to add a branch to the tree-structured network and construct a corresponding Bell-type inequality. We explore our inequalities in several examples, in all of which we demonstrate strong violations from quantum theory.Introduction.-The milstone work of John Bell showed that quantum correlations arising between spatially separated observers can break the limitations of classical physics [1]. Studies of correlations predicted by quantum theory has been a key to understanding fundamental properties of the theory and has led to applications in information processing including random number generation [2], device-independent cryptography [3] and reduction of communication complexity [4].A Bell experiment considers a source emitting a physical system shared between a set of observers who can randomly chose to apply some local measurements. The standard two-particle Bell experiment is illustrated in Figure 1 a). The properties of the correlations arising between the outcomes of the observers in Bell experiments have been thoroughly studied [5] and may be considered fairly well understood. However, much less is known about the nature of correlations arising in more sophisticated network configurations beyond the Bell experiments. A network could in general involve many independent sources each emitting a physical system that is shared between some set of observers. Importantly, a single observer could be receiving subsystems of many independent physical systems originating from different sources.There are both conceptual and applied motivations for studying classical and quantum correlations in networks. Networks naturally generalize Bell experiments which makes them conceptually attractive. Also, the notion of classical correlations on a network leads to stronger constraints than those associated to standard Bell inequalities. However, these constraints also influence the strength of quantum correlations, which makes it interesting to study their comparative strength as opposed to standard Bell inequalities. Furthermore, networks are relevant in a variety of applications involving entanglement swapping experiments [6], entanglement percolation [7] and quantum repeaters protocols [8,9]. Quantum correlations in networks are interesting for the practical implementation of large scale quantum communication networks which is arguably one of the main goals of applied quantum information.