Virtual agents, to be expressive, not only need algorithms for displaying the subtleties of human behaviour, but also require environments and tools so that people can author them. Because powerful algorithms are sometimes difficult to author, a compromise has to be found between algorithmic sophistication and authorability. Our approach for providing expressive characters at the behavioural level is based on such a compromise. This paper provides a model for describing behaviours which is author focused, while enabling some interesting algorithmic features such as parallelism and inter-agent coordination. The model has been implemented and simulation results are displayed.