Networks are built from nodes connected by edges. Nodes are the 'atoms' of networks. Just as atoms can combine into molecules that have new properties, nodes combine into network motifs (Milo et al., 2002). Network motifs are recurrent arrangements of nodes and edges that have certain dynamic properties such as oscillations and multi-stability and can provide useful input-output relations. Network motifs are the building blocks of complex networks. However, it is unclear how these building-block patterns are combined in real networks and what dynamic properties can emerge from their combinations. To address this, we define hyper-motifs: multiple network motifs that are direclty joined in the network. We developed a computational method to identify hyper-motifs in real networks and found that each type of network (social, neuronal, electronic, gene regulatory etc.) is enriched in some hyper-motifs, while other hyper-motifs are under-represented compared to random networks. We use the hyper-motif framework to explore behavior and structure of complex networks in complex tissue processes such as embryonic development.