Task-network modelling approaches are widely used to understand and improve product development (PD) processes. The best-practice application of these tools is often presented prescriptively. This paper proposes that best-practice, and academic understanding, of PD process modelling can be further developed through an empirical approach, in which modelling tools are taken as given and practice is analysed in terms of how industry stakeholders use these tools. We argue this inductive approach can result in process modelling guidelines which explicitly recognise the challenges of modelling in industry. We thus analyse four cases of process modelling at Rolls-Royce plc and find that: 1) modelling is a social process of knowledge-creation; 2) effective modelling requires selection of the appropriate methods and tools at the right time; and 3) understanding the purpose of each interaction between participants in the modelling process helps to choose the best approach. To conclude we present guidelines to suggest which modelling approach should be used in which context, where context is described in terms of the purpose of a given modelling interaction.