Attack trees are used in cybersecurity analysis to give an analyst a view of all the ways in which an attack can be carried out. Attack trees can become large, and developing them by hand can be tedious and error-prone. In this paper the automated generation of attack trees is considered. The method proposed is based on a library of attack templates – parameterisable patterns of attacks such as denial of service or eavesdropping – and that also uses an abstract model of the network architecture under attack. A pseudocode implementation of the method is also presented. The example application given is from the automotive domain and using an architecture consisting of linked CAN networks – a network configuration found in virtually every current vehicle.