There is a long history of research to create capabilities that address the need for human behaviour representations in training simulations and other M&S application domains. In training, human behaviour models have applications as synthetic teammates and adversaries, but can also be used as a representation of the state of the trainee and as synthetic instructors to increase the effectiveness of the training enterprise as a whole. They are essential compo-nents for achieving the goals for training simulations and for Live, Virtual, Constructive (LVC) training, including affordability, availability, and credibility. Over the last two decades, numerous formalisms and architectures for modelling cognition, performance, and other relevant characteristics of the human being have emerged, and the ca-pabilities and applications have expanded dramatically. However, models vary along many dimensions, including fidelity, application domain, underlying modelling formalisms, and behavioural repertoire. This diversity leads to critical challenges with respect to interoperability and reuse, in particular the integration of component models into a comprehensive behaviour model, and the integration of behaviour models into simulation environments. The chal-lenges are further complicated by a lack of standards for human behaviour modelling, leading to brittle models, lack of reusability, and increased costs driven by the requirements of model integration and reengineering. In this paper, we discuss the need for human behaviour modelling, its role in supporting affordable, available, and credible train-ing experiences in simulation and LVC environments, and propose a reference architecture to enable interoperabil-ity standards that support a variety of models serving a diverse set of purposes, both within and beyond the training domain. The authors represent a North Atlantic Treaty Organization (NATO) Modelling and Simulation Group (NMSG) activity focused on developing a baseline reference architecture and interoperability standards for human behaviour modelling to facilitate the creation and integration of human behaviour representations into simulation.