Research summary: Innovation requires inventors to have both new knowledge and the ability to combine and configure knowledge (i.e., combinatory knowledge), and such knowledge may flow through networks. We argue that both combinatory knowledge and new knowledge are accessed through collaboration networks, but that inventors' abilities to access such knowledge depends on its location in the network. Combinatory knowledge transfers from direct contacts, but not easily from indirect contacts. In contrast, new knowledge transfers from both direct and indirect contacts, but is far more likely to be new and useful when it comes from indirect contacts. Exploring knowledge flows in 69,476 patents and 89,930 unique inventors reveals evidence that combinatory knowledge from direct contacts and new knowledge from indirect contacts significantly affects innovative performance. Managerial summary: Inventors often combine ideas to create innovations. To do this, they need ideas to combine and they need the ability to combine those ideas. Inventors can get ideas to combine as well as the ability to combine ideas through prior co‐workers. Prior co‐workers can share ideas that may be relevant for the inventor's project and can tell the inventor about other things that other people are working on, especially people the inventor may not know. This can help inventors easily learn about ideas from friends‐of‐friends. The ability to combine ideas, however, is much harder to pass on. Prior co‐workers must carefully work with the inventor to teach him or her the complex processes of combining ideas. This means that it is very hard to learn how to combine knowledge from a friend‐of‐a‐friend, but it may be possible to learn from prior co‐workers. We explore this phenomenon in the social relationships of software inventors. Copyright © 2015 John Wiley & Sons, Ltd.